Publications

Dissemination (including electronic) of all results must acknowledge EU funding. See below for details.

Marco Viceconti and his team at the Insigneo Institute, part of the University of Sheffield have collated and written a report on the CompBioMed solutions and end-user engagement of our Core Partners. We are currently in the process of extending this to our Associate Partners

Preprints

 

Title Citation Summary
A Multi-Component, Multi-Physics Computational Model for Solving Coupled Cardiac Electromechanics and Vascular Haemodynamics Sharp C. Y. Lo, Alberto Zingaro, Jon W. S. McCullough, Xiao Xue, Mariano Vázquez, Peter V. Coveney, arXiv (2024) DOI: 10.48550/arXiv.2411.11797 This work introduces a novel computational framework coupling a 3D heart electromechanical model with a 3D vascular blood flow model to reflect the integrated dynamics of the circulatory system. Using a partitioned coupling scheme, the models run independently while exchanging data via intermediate files, facilitating collaboration between teams with differing expertise. Validation with two distinct solvers demonstrates reliable performance and negligible computational overhead compared to standalone simulations. Simulations reveal that incorporating vascular flow significantly impacts cardiac muscle displacement predictions, underscoring the model’s value in advancing virtual human models and digital twin technologies.
Towards Enabling Cardiac Digital Twins of Myocardial Infarction Using Deep Computational Models for Inverse Inference Li, Lei; Camps, Julia; Zhinuo; Wang; Banerjee, Abhirup; Beetz, Marcel; Rodriguez, Blanca; Grau, Vicente, arXiv (2024) DOI: 10.48550/arxiv.2307.04421 Exploring myocardial tissue properties inference from electrocardiogram (ECG) data within a cardiac digital twin (CDT) platform for personalized myocardial infarction (MI) diagnosis and treatment planning. Integrating multi-modal data enhances accuracy. A sensitivity analysis investigates infarct characteristics’ impact on ECG signals, informing a deep computational model development to infer infarct location and distribution. The model achieves promising results in infarct localization, offering clinical application potential. Insights into infarct characteristics’ relationship with electrophysiological features are provided, with planned code release upon publication acceptance.
3D Shape-Based Myocardial Infarction Prediction Using Point Cloud Classification Networks Beetz, M; Yang, Y; Banerjee, A; Li, L; Grau, V, arXiv (2023) DOI: 10.48550/arxiv.2307.07298 Myocardial infarction (MI) is a prevalent cardiovascular disease, often diagnosed using single-valued imaging biomarkers that only approximate the heart’s complex 3D structure and physiology, limiting understanding and prediction of MI outcomes. This study explores the use of complete 3D cardiac shapes represented as point clouds for enhanced MI detection. A fully automatic multi-step pipeline is proposed, involving 3D cardiac surface reconstruction followed by point cloud classification using geometric deep learning. The method leverages recent advancements to enable direct and efficient multi-scale learning on high-resolution cardiac anatomy models. Evaluation on 1068 UK Biobank subjects shows improvements of approximately 13% in prevalent MI detection and around 5% in incident MI prediction compared to clinical benchmarks. Additionally, the study analyzes the role of each ventricle and cardiac phase in 3D shape-based MI detection and visually examines morphological and physiological patterns associated with MI outcomes.
Anatomical basis of sex differences in human post-myocardial infarction ECG phenotypes identified by novel automated torso-cardiac 3D reconstruction Hannah J. Smith, Blanca Rodriguez, Yuling Sang, Marcel Beetz, Robin Choudhury, Vicente Grau, Abhirup Banerjee, arXiv (2023) DOI: 10.48550/arXiv.2312.13976 The electrocardiogram (ECG) is a common tool in cardiology, but its interpretation is complicated by anatomical differences. A new automated computational pipeline allows quantification of torso-ventricular anatomy metrics from magnetic resonance imaging (MRI), comparing them to ECG characteristics. Sex and myocardial infarction (MI) differences are explored using data from 1051 healthy and 425 post-MI subjects from UK Biobank. Smaller ventricles in females explain around 50% of shorter QRS durations compared to males, contributing to lower STJ amplitudes due to different ventricular positions. In females, torso-ventricular anatomy, particularly influenced by larger BMI, more strongly influences T wave amplitude reductions and left-deviated R axis angles post-MI compared to males. Consequently, the female MI phenotype may be less indicative of pathology, with baseline STJ amplitudes and QRS durations further from clinical thresholds. Quantifying anatomical sex differences and their impact on ECG in health and disease is crucial to mitigate clinical sex bias.
Clinical phenotypes in acute and chronic infarction explained through human ventricular electromechanical modelling and simulations Xin Zhou, Zhinuo Jenny Wang, Julia Camps, Jakub Tomek, Alfonso Santiago, Adria Quintanas, Mariano Vazquez, Marmar Vaseghi, Blanca Rodriguez, bioRxiv (2023) DOI: 10.1101/2022.02.15.480392 The study aims to provide a mechanistic understanding of clinical phenotypes in acute and chronic myocardial infarction (MI), linking ionic remodeling to electrocardiogram (ECG) changes and ejection fraction (EF) using human electromechanical modeling and simulation. A framework is developed and validated with experimental and clinical data. Simulations with varying degrees of scar and border zone ionic remodeling reproduce clinical phenotypes post-MI, explaining T-wave inversion and Brugada phenocopy in acute MI and tall T-waves in chronic MI. Ionic remodeling impacts EF through calcium transient amplitude inhibition, but EF at resting heart rate is insensitive to repolarization heterogeneity. Multi-scale modeling integrates data to elucidate electromechanical disease mechanisms in MI, highlighting the importance of ionic remodeling and repolarization dispersion in post-MI phenotypes. T-wave and QT abnormalities are superior indicators of repolarization heterogeneities compared to EF post-MI.
Digital Twinning of the Human Ventricular Activation Sequence to Clinical 12-lead ECGs and Magnetic Resonance Imaging Using Realistic Purkinje Networks for in Silico Clinical Trials Julia Camps, Lucas Arantes Berg, Zhinuo Jenny Wang, Rafael Sebastian, Leto Luana Riebel, Ruben Doste, Xin Zhou, Rafael Sachetto, James Coleman, Brodie Lawson, Vicente Grau, Kevin Burrage, Alfonso Bueno-Orovio, Rodrigo Weber, Blanca Rodriguez, arXiv (2023) DOI: 10.48550/arXiv.2306.13740 Introducing a novel digital twinning pipeline for cardiac in silico clinical trials. By integrating subject-specific clinical data, including 12-lead electrocardiograms and magnetic resonance recordings, we efficiently generate and integrate Purkinje networks into multiscale biventricular models. Key features include personalized Purkinje network generation based on ECG characteristics and translation to detailed biophysical models. Simulations align closely with clinical data in healthy and hypertrophic cardiomyopathy patients, demonstrating the pipeline’s effectiveness in generating clinically-consistent ECGs. Additionally, our method accounts for regional differences in Purkinje density, impacting conduction velocities and coupling effects in the model.
Equivariant Graph Neural Networks for Toxicity Prediction Julian Cremer; Leonardo Medrano Sandonas; Alexandre Tkatchenko; Djork-Arné Clevert; Gianni De Fabritiis, ChemRxiv (2023) DOI: 10.26434/chemrxiv-2023-9kb55 Toxicity prediction is vital in drug discovery, often achieved through machine learning (ML) models using molecular representations like fingerprints or SMILES strings. However, molecules exist in 3D space, prompting exploration of 3D representations for toxicity prediction. We investigate equivariant graph neural networks (EGNNs), specifically equivariant transformers (ET), for toxicity prediction. Our study utilizes 11 toxicity datasets and TorchMD-NET’s ET model, achieving competitive accuracies compared to 2D graph-based models. While conformational analysis yields minimal impact, incorporating physicochemical properties like total energy doesn’t enhance predictive performance. Attention weight analysis aids in explaining predictions, underscoring the importance of 3D geometry in toxicity prediction models.
In silico Evaluation of Cell Therapy in Acute versus Chronic Infarction Leto Luana Riebel, Zhinuo Jenny Wang, Hector Martinez-Navarro, Cristian Trovato, Julia Camps, Lucas Arantes Berg, Xin Zhou, Ruben Doste, Rafael Sachetto Oliveira, Rodrigo Weber dos Santos, Jacopo Biasetti, Blanca Rodriguez, bioRxiv (2023) DOI: 10.1101/2023.12.11.570780 Human-based modelling and simulation guide medical therapy development. We explore cell therapy’s impact on acute versus chronic myocardial infarction (MI), considering cell heterogeneity, scar size, and the Purkinje system. Our simulations, validated against experimental and clinical data, reveal that cell heterogeneity and chronic MI foster spontaneous beats, leading to re-entrant arrhythmias. These arrhythmias, exacerbated by impaired Purkinje-myocardium coupling, large scars, and acute infarction, underscore the importance of uniform cell populations and healthy Purkinje-myocardium coupling in mitigating arrhythmic risks.
Modeling 3D cardiac contraction and relaxation with point cloud deformation networks Beetz, Marcel; Banerjee, Abhirup; Grau, Vicente, arXiv (2023) DOI: 10.48550/arxiv.2307.10927 Global single-valued biomarkers like ejection fraction offer limited insight into the true 3D cardiac deformation process, hindering understanding of cardiac mechanics in health and disease. To address this, we introduce the Point Cloud Deformation Network (PCD-Net), a novel geometric deep learning approach to model 3D cardiac contraction and relaxation. Utilizing point cloud-based deep learning, PCD-Net enables efficient multi-scale feature learning directly on 3D point cloud representations of cardiac anatomy. Evaluation on a large dataset from the UK Biobank study demonstrates accurate predictions with Chamfer distances below image resolution. Clinical metrics between predicted and ground truth populations are similar, and PCD-Net effectively captures subpopulation-specific differences between normal subjects and myocardial infarction (MI) patients. Moreover, learned 3D deformation patterns outperform clinical benchmarks in prevalent MI detection, incident MI prediction, and MI survival analysis.
Prediction of pyrazinamide resistance in Mycobacterium tuberculosis using structure-based machine learning approaches. Carter JJ, Walker TM, Waker AS, Whitfield M, Morlock GP, Lynch CI, Adlard D, Peto TEA, Posey JE, Crook DW, Fowler PW, bioRxiv (2023) DOI: 10.1101/518142 Pyrazinamide resistance poses challenges in tuberculosis treatment due to difficulties in antibiotic susceptibility testing. This study addresses this issue by curating a dataset of 664 non-redundant mutations in pncA, a gene associated with pyrazinamide resistance. Three machine learning models, leveraging protein structural, chemical, and sequence-based features, were trained to predict resistance. The best-performing model achieved a sensitivity of 80.2% and specificity of 76.9%. Clinical evaluation on 4,027 samples highlighted the utility of machine learning in enhancing resistance prediction, identifying novel mutations, and showcasing its potential application in other drug resistance contexts.
TensorNet: Cartesian Tensor Representations for Efficient Learning of Molecular Potentials Simeon, Guillem; de Fabritiis, Gianni, arXiv (2023) DOI: 10.48550/arxiv.2306.06482 TensorNet introduces an innovative O(3)-equivariant message-passing neural network architecture for molecular systems representation, utilizing Cartesian tensor atomic embeddings. By simplifying feature mixing through matrix product operations and cost-effective tensor decomposition into rotation group irreducible representations, TensorNet achieves state-of-the-art performance with fewer parameters compared to higher-rank spherical tensor models. Even with a single interaction layer, TensorNet reduces computational costs for small molecule potential energies. Additionally, it accurately predicts vector and tensor molecular quantities alongside potential energies and forces. This framework presents a promising avenue for designing advanced equivariant models in molecular research.
Fluid-structure interaction analysis of eccentricity and leaflet rigidity on thrombosis biomarkers in bioprosthetic aortic valve replacements David Oks; Mariano Vázquez; Guillaume Houzeaux; Constantine Butakoff; Cristóbal Samaniego, bioRxiv (2022) DOI: 10.1101/2022.01.06.475272 Presenting the inaugural 2-way fluid-structure interaction (FSI) computational model, this study investigates the influence of aortic annulus eccentricity on cardiac bioprostheses’ function and thrombogenic risk. Higher eccentricities correlate with decreased geometric orifice areas (GOAs) and increased transvalvular pressure gradients (TPGs), consistent with experimental data. Elevated thrombus formation risk is noted in sinus of Valsalva regions with peak residence time and shear rate for eccentric configurations. Furthermore, the model evaluates the impact of leaflet rigidity on performance and thrombogenicity, indicating heightened TPGs and thrombogenic risk with rigid leaflets. Validated against benchmarks and experimental data, this model, executed on high-performance computing, provides a valuable tool for device manufacturers and clinicians.
In-silico clinical trial using high performance computational modeling of a virtual human cardiac population to assess drug-induced arrhythmic risk J. Aguado-Sierra, C. Butakoff, R. Brigham, A. Baron, G. Houzeaux, J. M. Guerra, F. Carreras, D. Filgueiras-Rama, P. A. Iaizzo, T. L. Iles, M. Vazquez, medRxiv (2022) DOI: 10.1101/2021.04.21.21255870 Cardiotoxicity remains a global health concern, necessitating safe drug access. We propose an in-silico clinical trial workflow, simulating 3D biventricular human populations’ responses to potentially cardiotoxic drugs. Gender-specific ionic channel characteristics are replicated to assess arrhythmic risk under drug influence via electrophysiology simulations. Hydroxychloroquine and Azithromycin exemplify the method, validated against swine heart experiments, showing similar clinical trial outcomes. Results suggest transmural action potential prolongation as an arrhythmia mechanism, aligning with clinical data. This workflow offers rapid drug risk assessment, demonstrating diverse phenotypes’ distinct arrhythmogenic outcomes, crucial for drug safety evaluation.
TorchMD-NET: Equivariant Transformers for Neural Network based Molecular Potentials Thölke, Philipp; De Fabritiis, Gianni, arXiv (2022) DOI: 10.48550/arxiv.2202.02541 This study presents TorchMD-NET, a novel equivariant transformer (ET) architecture for predicting quantum mechanical properties with improved accuracy and computational efficiency. Outperforming state-of-the-art models on MD17, ANI-1, and QM9 targets, TorchMD-NET offers valuable insights into the black box predictor through attention weight analysis. Differences in learned representation between conformers and conformations from molecular dynamics or normal modes are elucidated. Additionally, the study emphasizes the significance of datasets including off-equilibrium conformations for evaluating molecular potentials. These findings advance the prediction of quantum mechanical properties, addressing longstanding challenges in accuracy-speed trade-offs.
Enhanced single-node boundary condition for the Lattice Boltzmann Method Marson, Francesco; Thorimbert, Yann; Latt, Jonas; Chopard, Bastien, arXiv (2020) DOI: 10.48550/arxiv.2009.04604 We introduce a novel approach to implementing Dirichlet boundary conditions for complex shapes in the lattice Boltzmann method, utilizing data from a single node. This method achieves second-order convergence for velocity fields and demonstrates similar or superior accuracy to established boundary conditions for curved walls. It is adaptable for simulating moving rigid objects or immersed surfaces with or without prescribed motion. By limiting interpolation information to the boundary’s proximity, our approach enhances the bounce-back rule, providing a more efficient and accurate solution for complex geometries.
Anomalous Platelet Transport & Fat-Tailed Distributions C. Kotsalos, K. Z. Boudjeltia, R. Dutta, J. Latt, B. Chopard, (2020), DOI: arXiv:2006.11755 The transport of platelets in blood is commonly assumed to obey an advection-diffusion equation. Here we propose a disruptive view, by showing that the random part of their velocity is governed by a fat-tailed probability distribution, usually referred to as a Lévy flight. Although for small spatio-temporal scales, it is hard to distinguish it from the generally accepted “red blood cell enhanced” Brownian motion, for larger systems this effect is dramatic as the standard approach may underestimate the flux of platelets by several orders of magnitude, compromising in particular the validity of current platelet function tests.

2024

Title Citation Summary
Equilibrium and Nonequilibrium Ensemble Methods for Accurate, Precise and Reproducible Absolute Binding Free Energy Calculations A. P. Bhati, S. Wan, P. V. Coveney, Journal of Chemical Theory and Computation (2024), DOI: 10.1021/acs.jctc.4c01389 This study advances the application of ensemble alchemical methods for calculating absolute binding free energies (ABFEs) in protein-ligand complexes, a crucial task for drug discovery. By analyzing 219 complexes, the results demonstrate high accuracy (<1 kcal/mol) and robust statistical reproducibility. Equilibrium methods outperform non-equilibrium approaches, being more accurate, efficient, and computationally economical, with a simpler protocol suitable for large-scale applications. The study also highlights the non-normal nature of free energy distributions and its implications. A recommended protocol is provided, enabling thousands of ABFE calculations within hours on modern exascale computing platforms, offering transformative potential for drug discovery pipelines.
The First Virtual Human Global Summit: Prepublication Meeting Report E. A. Stahlberg, K. K. Van Dam, P. V. Coveney et. al., U.S. Department of Energy, Office of Scientific and Technical Information (2024), OSTI ID: 2428904 The prepublication report for the 2023 Virtual Human Global Summit details a collaborative event focused on advancing patient-focused precision medicine through medical digital twins. Organized by leading institutions, the Summit gathered over 80 global leaders from diverse sectors, including cancer research, industry, and community health, to discuss the current state, challenges, and opportunities in digital twin technology. The event featured sessions on research, infrastructure, clinical application, and sustainability, with examples of digital twin applications in cancer, diabetes, and sepsis. The report captures insights from the Summit, offering guidance and highlighting key challenges and industry perspectives.
Uncertainty quantification of the lattice Boltzmann method focussing on studies of human-scale vascular blood flow J. W. S. McCullough, P. V. Coveney, Scientific Reports (2024), DOI: 10.1038/s41598-024-61708-w Uncertainty quantification is becoming a key tool to ensure that numerical models can be sufficiently trusted to be used in domains such as medical device design. Demonstration of how input parameters impact the quantities of interest generated by any numerical model is essential to understanding the limits of its reliability. With the lattice Boltzmann method now a widely used approach for computational fluid dynamics, building greater understanding of its numerical uncertainty characteristics will support its further use in science and industry. In this study we apply an in-depth uncertainty quantification study of the lattice Boltzmann method in a canonical bifurcation geometry that is representative of the vascular junctions present in arterial and venous domains. Our work provides insights into how input parameters and boundary conditions impact the velocity and pressure distributions calculated in a simulation and can guide the choices of such values when applied to vascular studies of patient specific geometries. This study also demonstrates how open-source toolkits for validation, verification and uncertainty quantification can be applied to numerical models deployed on high-performance computers without the need for modifying the simulation code itself. Such an ability is key to the more widespread adoption of the analysis of uncertainty in numerical models by significantly reducing the complexity of their execution and analysis.
High Performance Computing for Drug Discovery and Biomedicine Alexander Heifetz, Shunzhou Wan, Peter V. Coveney, Marco Verdicchio, Carlos Teijeiro Barjas, Gavin J. Pringle, Simon Arsène, Yves Parès, Eliott Tixier, Solène Granjeon-Noriot, Bastien Martin, Lara Bruezière et al., Lionel Colliandre, Christophe Muller, Vladimir Joseph Sykora, Bhushan Bonde, Pratik Patil, Bhaskar Choubey, Bhushan Bonde, Tim James, Holger Hennig, J. Charles G. Jeynes, Tim James, Matthew Corney, Anna M. Herz, Tahsin Kellici, Inaki Morao, Julien Michel, Martin Kotev, Constantino Diaz Gonzalez, Martin Kotev, Constantino Diaz Gonzalez, Reuben L. Martin, Alexander Heifetz, Mike J. Bodkin, Andrea Townsend-Nicholson, Jazmin Aguado-Sierra, Renee Brigham, Apollo K. Baron, Paula Dominguez Gomez, Guillaume Houzeaux, Jose M. Guerra et al., Zainab Altai, Erica Montefiori, Xinshan Li, Gabor Zavodszky, Christian Spieker, Benjamin Czaja, Britt van Rooij, Remy Petkantchin, Franck Raynaud, Karim Zouaoui Boudjeltia, Bastien Chopard, Ivan Pribec, Stephan Hachinger, Mohamad Hayek, Gavin J. Pringle, Helmut Brüchle, Ferdinand Jamitzky et al., Andrea Townsend-Nicholson, Springer (2024) DOI: 10.1007/978-1-0716-3449-3 “High Performance Computing for Drug Discovery and Biomedicine” explores the fusion of high-performance computing (HPC) technologies with computational drug discovery (CDD) and biomedicine. The volume is divided into two sections: the first focuses on CDD approaches empowered by HPC, such as knowledge graphs, natural language processing (NLP), and virtual screening platforms. It also explores techniques like alchemical free energy workflows and molecular dynamic simulations adapted for HPC. Additionally, it discusses the potential of cloud computing in drug discovery. The second section covers computational algorithms and workflows for biomedicine, including assessing drug-induced arrhythmic risk, digital patient applications, virtual human simulations, and blood flow modeling for stroke treatments. Each chapter in this Methods in Molecular Biology series provides introductions to the topics, lists necessary software and tools, offers reproducible modeling protocols, and includes troubleshooting tips. This comprehensive guide caters to a wide audience, from computer scientists to drug designers, providing insights into current possibilities, challenges, and future directions in HPC-based technologies for drug discovery and biomedicine.
Segment anything model for medical images? Yuhao Huang and Xin Yang and Lian Liu and Han Zhou and Ao Chang and Xinrui Zhou and Rusi Chen and Junxuan Yu and Jiongquan Chen and Chaoyu Chen and Sijing Liu and Haozhe Chi and Xindi Hu and Kejuan Yue and Lei Li and Vicente Grau and Deng-Ping Fan and Fajin Dong and Dong Ni, Medical Image Analysis (2024) DOI: 10.1016/j.media.2023.103061 The Segment Anything Model (SAM), renowned for its success in general image segmentation, faces unique challenges in medical image segmentation (MIS) due to complex modalities and anatomical structures. To assess SAM’s performance in MIS, a comprehensive dataset, COSMOS 1050K, comprising 18 modalities and 84 objects, was compiled from 53 open-source datasets. SAM exhibited notable performance variations across different objects and modalities, with larger ViT-H models generally outperforming ViT-B. SAM benefited from manual hints and aided human annotation efficiency. However, its sensitivity to randomness and performance compared to interactive methods varied. Finetuning SAM on specific medical tasks yielded performance enhancements, demonstrating its potential and areas for improvement in MIS applications.

2023

Title Citation Summary
High resolution simulation of basilar artery infarct and flow within the circle of Willis J. W. S. McCullough, P. V. Coveney, Scientific Reports (2023), DOI: 10.1038/s41598-023-48776-0 On a global scale, cerebro-and cardiovascular diseases have long been one of the leading causes of death and disability and their prevalence appears to be increasing in recent times. Understanding potential biomarkers and risk factors will help to identify individuals potentially at risk of suffering an ischemic stroke. However, the widely variable construction of the cerebral vasculature makes it difficult to provide a specific assessment without the knowledge of a patient’s physiology. In this paper we use the 3D blood flow simulator HemeLB to study flow within three common structural variations of the circle of Willis during and in the moments after a blockage of the basilar artery. This tool, based on the lattice Boltzmann method, allows the 3D flow entering the basilar artery to be finely controlled to replicate the cessation of blood feeding this particular vessel-we demonstrate this with several examples including a sudden halt to flow and a gradual loss of flow over three heartbeat cycles. In this work we start with an individualised 3D representation of a full circle of Willis and then construct two further domains by removing the left or right posterior communicating arteries from this geometry. Our results indicate how, and how quickly, the circle of Willis is able to redistribute flow following such a stroke. Due to the choice of infarct, the greatest reduction in flow was observed in the posterior cerebral arteries where flow was reduced by up to 70% in some cases. The high resolution domains used in this study permit the velocity magnitude and wall shear stress to be analysed at key points during and following the stroke. The model we present here indicates how personalised vessels are required to provide the best insight into stroke risk for a given individual.
Large-Scale Molecular Dynamics Elucidates the Mechanics of Reinforcement in Graphene-Based Composites J. L. Suter, M. Vassaux, P. V. Coveney, Advanced Materials (2023) DOI: 10.1002/adma.202302237 Using very large-scale classical molecular dynamics we examine the mechanics of nano-reinforcement of graphene-based nanocomposites. Our simulations show that significant quantities of large, defect-free and predominantly flat graphene flakes are required for successful enhancement of materials properties in excellent agreement with experimental and proposed continuum shear-lag theories. The critical length for enhancement is approximately 500nm and 300nm for graphene and GO respectively. The reduction of Young’s modulus in GO results in a much smaller enhancement of the composite’s Young’s modulus. The simulations reveal that the flakes should be aligned and planar for optimal reinforcement. Undulations substantially degrade the enhancement of materials properties.
Artificial Intelligence for Science: Chapter 21: Big AI: Blending Big Data with Big Theory to Build Virtual Humans Peter Coveney and Roger Highfield, World Scientific (2023) DOI: 10.1142/9789811265679_0021 Computational scientists envision a future where the health of individual patients can be forecasted using mathematical models of the body fueled by personalized data, akin to weather forecasting systems. Teams worldwide are integrating data and models to simulate cells, tissues, and organs on high-performance computers, aiming to create “virtual humans” for next-generation healthcare. Artificial intelligence (AI), particularly machine learning (ML), plays a crucial role in this endeavor, notably in computer-aided drug discovery. However, while ML has its successes, there are limitations to its capabilities. Combining AI with mechanistic understanding, termed “Big AI,” offers a more potent approach. This involves an iterative cycle where AI hypotheses are tested in physics-based simulations, and the results inform AI training. This concept has gained traction across various fields, including earth systems, climate science, and materials science.
Ensemble-Based Approaches Ensure Reliability and Reproducibility S. Wan, A. Bhati, A. Wade, Alexander,P. V. Coveney, Journal of Chemical Information and Modeling (2023) DOI: 10.1021/acs.jcim.3c01654 Increasingly, there is a widespread acknowledgment of the necessity for ensemble-based approaches to ensure reliability, accuracy, and precision in molecular dynamics calculations. The focus of this discussion is on addressing a frequently raised question: what constitutes the optimal approach for conducting ensemble simulations to calculate quantities of interest?
Comparison of Equilibrium and Non-Equilibrium Approaches for Relative Binding Free Energy Predictions S. Wan, A. Bhati, P. V. Coveney, Journal of Chemical Theory and Computation (2023) DOI: 10.1021/acs.jctc.3c00842 Alchemical relative binding free energy calculations have become increasingly important in drug optimization, where a series of related compounds are evaluated for their binding affinities to a protein to refine potential drugs. While equilibrium thermodynamics-based methods are well-studied, a newer nonequilibrium approach has been touted as superior. However, these claims lack sufficient scrutiny of both methods’ foundations and reliability. In this study, we compare the two approaches across a large dataset comprising over 500 ligand transformations involving more than 300 ligands binding to 14 diverse protein targets. Ensemble methods are crucial for quantifying uncertainty in these calculations, ensuring the nonequilibrium approach remains within its valid domain. We find that, with the application of ensemble methods, the nonequilibrium method can achieve accuracy and precision comparable to the equilibrium approach. However, it comes with increased computational complexity and longer runtime. Additionally, inadequate transition length in nonequilibrium calculations may necessitate a complete rerun, significantly elevating computational costs. Our findings offer recommendations for the reliable implementation of nonequilibrium approaches in relative binding free energy calculations for ligand-protein systems.
Virtual You: How Building Your Digital Twin Will Revolutionize Medicine and Change Your Life P. V. Coveney and R. R. Highfield, Princeton University Press (2023) DOI: 10.1515/9780691223407 “Virtual You” provides a comprehensive overview of global scientific endeavors to construct digital replicas of human beings, spanning from cells and tissues to entire organs and bodies. These virtual counterparts herald a new era of personalized medicine, where your digital twin can forecast disease risk, engage in virtual drug trials, offer insights into optimal diet and lifestyle changes, and aid in identifying therapies for improved well-being and longevity. However, significant challenges lie ahead. Authors Peter Coveney and Roger Highfield outline a five-step process required to construct a functional digital replica of a person. Through captivating prose, they guide readers through the intricacies of the human body, exploring cutting-edge scientific and technological breakthroughs—from multiscale modeling to advanced computing—that will bring the concept of “virtual you” to fruition. Simultaneously, they delve into the ethical considerations inherent in achieving predictive medicine.
Development and performance of a HemeLB GPU code for human-scale blood flow simulation I. Zacharoudiou, J. W. S. McCullough, P. V. Coveney, Computer Physics Communications (2023) DOI: 10.1016/j.cpc.2022.108548 In recent years, high-performance computers (HPC) increasingly feature heterogeneous architectures, often integrating GPU accelerators. These accelerators, either in dedicated partitions or integral to all compute nodes, contribute significantly to a machine’s compute performance. Consequently, updating HPC codes to execute on accelerator hardware is imperative. This paper introduces a GPU implementation of the 3D blood flow simulation code HemeLB, developed using CUDA C++. By harnessing NVIDIA GPU hardware, the implementation achieves substantial performance enhancements compared to its CPU-only counterpart, while preserving excellent strong scaling characteristics. As HPC advances towards the exascale era, the study uses HemeLB as a case study to address the challenges users may encounter when deploying applications on upcoming exascale machines.
A Blood Flow Modeling Framework for Stroke Treatments Remy Petkantchin, Franck Raynaud, Karim Zouaoui Boudjeltia and Bastien Chopard, High Performance Computing for Drug Discovery and Biomedicine (2023) DOI: 10.1007/978-1-0716-3449-3_17 Circulatory models offer promise in mitigating the impact of stroke, yet determining hemodynamics conditions and simulating porous media remain challenging. This chapter introduces a validated open-source lattice-Boltzmann numerical framework designed for such issues. Key features include an algorithm for pressure boundary condition imposition and simulation of porous media permeability using a method developed by Walsh et al. The framework’s capabilities are demonstrated through a thrombolysis model, highlighting its flexibility and potential for stroke research.
Binding-and-Folding Recognition of an Intrinsically Disordered Protein Using Online Learning Molecular Dynamics Pablo Herrera-Nieto; Adrià Pérez; Gianni De Fabritiis, Journal of Chemical Theory and Computation (2023) DOI: 10.1021/acs.jctc.3c00008 Intrinsically disordered proteins play vital roles in various biological processes by folding upon binding to other proteins. However, the dynamics of coupled folding and binding processes remain poorly understood at the atomic level. A key question is whether folding occurs prior to or after binding. In this study, we employ a novel, unbiased, high-throughput adaptive sampling approach to elucidate the binding and folding dynamics between the disordered transactivation domain of c-Myb and the KIX domain of the CREB-binding protein. Our reconstructed long-term dynamical process reveals the binding of a short stretch of amino acids on c-Myb as a folded α-helix. Specifically, leucine residues, particularly Leu298-Leu302, establish initial native contacts that facilitate the binding and folding of the remaining peptide, involving a combination of conformational selection on the N-terminal region and induced fit of the C-terminal.
Computational Biomedicine (CompBioMed) Centre of Excellence: Selected Key Achievements Pringle, GJ, High Performance Computing for Drug Discovery and Biomedicine (2023) DOI: 10.1007/978-1-0716-3449-3_3 CompBioMed, a Centre of Excellence for High Performance Computing Applications, funded by the European Commission’s Horizon 2020 program, operates from October 1, 2017, to April 1, 2024. Its mission is to develop computer-based tools for simulating the human body in health and disease. The author provides a general overview of CompBioMed’s objectives and then offers personal insights into key achievements. These include successful collaborations between industry and academia, the production of two IMAX short films, initiatives for training to promote HPC culture among biomedical practitioners, a free service for porting and tuning biomedical applications to HPC, facilitating access to future supercomputers for surgical simulations, and enhancing credibility for biomedical simulations through FDA endorsement.
Cost-effectiveness analysis of CT-based finite element modelling for osteoporosis screening in secondary fracture prevention: an early health technology assessment in the Netherlands Li J., Viceconti M., Li X., Bhattacharya P., Naimark D., Osseyran A., MDM Policy & Practice (2023) DOI: 10.1177/23814683231202993 A new osteoporosis screening method called CT2S, using CT scans to assess bone strength, was compared to existing methods in Dutch postmenopausal women. Results show CT2S is cost-effective, especially for ages 70 and above, with potential savings and better outcomes compared to no screening or treating everyone. Adjusting treatment types and screening intervals can improve cost-effectiveness. This suggests CT2S as a promising tool for preventing secondary fractures.
Deep Computational Model for the Inference of Ventricular Activation Properties Li, Lei; Camps, Julia; Banerjee, Abhirup; Beetz, Marcel; Rodriguez, Blanca; Grau, Vicente, International Workshop on Statistical Atlases and Computational Models of the Heart (2023) DOI: 10.1007/978-3-031-23443-9_34 Patient-specific cardiac models are vital for personalized medicine and virtual clinical trials. Digital twins offer tailored cardiac function insights, aiding diagnosis and therapy planning. Existing workflows lack efficiency and accuracy. A proposed deep learning model combines anatomical and electrophysiological data to predict ventricular activation properties, crucial for intervention guidance. Simulated electrocardiograms train the model, considering patient-specific information. Evaluation reveals promising results with rapid computation, suggesting potential for enhancing cardiac interventions.
Deep Learning-Based Emulation of Human Cardiac Activation Sequences Bertrand, Ambre; Camps, Julia; Grau, Vicente; Rodriguez, Blanca; , International Conference on Functional Imaging and Modeling of the Heart (2023) DOI: 10.1007/978-3-031-35302-4_22 The aim of precision cardiology with digital twins is to merge patient data and expert knowledge for personalized treatment plans. However, current methods for simulating cardiac electrophysiology are computationally intensive. This study explores a U-Net-based model to speed up simulations while maintaining accuracy. By standardizing anatomical and electrophysiological data, the model predicts cardiac activation times swiftly. Results show it’s up to 500 times faster than traditional methods with comparable accuracy, offering potential for personalized simulations in large patient groups.
Effect of muscle forces on femur during level walking using a virtual population of older women Z. Altai, E. Montefiori, X. Li, Methods in Molecular Biology (2023) DOI: 10.1007/978-1-0716-3449-3_15 Aging increases the risk of muscle and bone disorders, affecting mobility and quality of life. Past studies often treated muscles and bones separately with generic data, neglecting individual differences in the elderly. This study introduces a novel approach, coupling personal-specific musculoskeletal and finite element models to study femoral loading during walking. Lower limb muscle variations were analyzed using a virtual population, revealing impacts on femur strain. Effective coupling across scales was achieved, shedding light on muscle-bone interaction in elderly women. Generating a virtual population offers a practical solution to represent anatomical variations without extensive data collection.
Efficient and Reliable Data Management for Biomedical Applications Pribec I, Hachinger S, Hayek M, Pringle GJ, Brüchle H, Jamitzky F, Mathias G, High Performance Computing for Drug Discovery and Biomedicine (2023) DOI: 10.1007/978-1-0716-3449-3_18 This chapter addresses the challenges and needs of modern Research Data Management (RDM) in biomedical applications within high-performance computing (HPC) contexts, emphasizing the FAIR data principles (Findable, Accessible, Interoperable, Reusable). It discusses data formats, publication platforms, annotation schemata, automated data management, and infrastructure in HPC centers. Additionally, it covers file transfer methods, EUDAT components, automated data movement in cross-center workflows, and ontology development for metadata structuring. Real-world examples, such as the CompBioMed project, demonstrate the implementation of these principles and tools. The LEXIS project presents a workflow-execution and data management platform, enhancing accessibility to HPC and Cloud Computing through user-friendly portals. Resilient workflows are ensured through checkpointing and data duplication, facilitating urgent computing on exascale platforms.
Enhanced optimization-based method for the generation of patient-specific models of Purkinje networks Lucas Arantes Berg, Bernardo Martins Rocha, Rafael Sachetto Oliveira, Rafael Sebastian, Blanca Rodriguez, Rafael Alves Bonfim de Queiroz, Elizabeth M. Cherry & Rodrigo Weber dos Santos, Scientific Reports (2023) DOI: 10.1038/s41598-023-38653-1 This work introduces an optimization-based method for generating patient-specific Purkinje networks, addressing their complex morphology. The method ensures accuracy in branch size, bifurcation angles, and Purkinje-ventricular-junction activation times. Evaluation on three biventricular meshes shows the method’s effectiveness. Monodomain simulations using recent cellular models confirm the realistic behavior of the generated networks. The results illustrate the capability of the new method to produce patient-specific Purkinje networks with controlled morphological metrics and specified local activation times, advancing our ability to model cardiac arrhythmias accurately.
High-Throughput Structure-Based Drug Design (HT-SBDD) Using Drug Docking, Fragment Molecular Orbital Calculations, and Molecular Dynamic Techniques. Martin RL, Heifetz A, Bodkin MJ, Townsend-Nicholson A, Methods in Molecular Biology (2023) DOI: 10.1007/978-1-0716-3449-3_13 Structure-based drug design (SBDD) emerges as a crucial tool for expedited and cost-effective lead drug discovery, aiming to replace traditional high-throughput screening (HTS) methods. This approach, known as “virtual screening,” leverages target protein structural data and vast databases of potential drug candidates. Computational techniques are then applied to identify candidates with high binding affinity and efficacy. High-throughput SBDD (HT-SBDD) is expected to enhance HTS success rates, currently around ~1%. This chapter explores high-throughput drug docking, fragment molecular orbital calculations, and molecular dynamics techniques, comparing traditional methods with recent SBDD advancements. Given the computational intensity of HT-SBDD, the role of high-performance computing (HPC) clusters in computational drug discovery is also discussed.
Image-based flow simulation of platelet aggregates under different shear rates Y. Hao, G. Závodszky, C. Tersteeg, M. Barzegari, and A. G. Hoekstra, PLOS Computational Biology (2023) DOI: 10.1371/journal.pcbi.1010965 The paper introduces a novel image-based computational model to simulate blood flow through and around platelet aggregates, crucial for understanding hemodynamics and platelet physiology. Utilizing microscopy images from in vitro experiments, the model captures aggregate microstructure via two imaging modalities. Platelet aggregates are represented as a porous medium, and their permeability is calculated using the Kozeny-Carman equation. The model analyzes blood flow velocity, shear stress, and kinetic forces on aggregates under various wall shear rates. Additionally, it evaluates agonist transport within aggregates using the local Péclet number. Results highlight the impact of shear rate and microstructure on agonist transport and identify zones of high kinetic forces at aggregate boundaries. The study also investigates the influence of shear rate and flow elongation on aggregate shapes. Overall, the model provides insights into platelet aggregate hemodynamics and physiology, facilitating predictions of aggregation and deformation under diverse flow conditions.
Influence of Myocardial Infarction on QRS Properties: A Simulation Study Li, Lei; Camps, Julia; Zhinuo; Wang; Banerjee, Abhirup; Rodriguez, Blanca; Grau, Vicente, International Conference on Functional Imaging and Modeling of the Heart (2023) DOI: 10.1007/978-3-031-35302-4_23 The interplay between structural and electrical changes post-myocardial infarction (MI) significantly influences arrhythmia. This study systematically investigates 17 post-MI scenarios, varying scar location, size, transmural extent, and conductive level, to understand their effects on QRS morphology. Using a Eikonal model, electrical activity propagation in the ventricles is simulated, and QRS scores are analyzed. Results suggest that QRS variations can identify MI and may allow inverse reconstruction of infarct regions. Sensitivity variations of QRS criteria for different MI scenarios provide insights for solving the inverse problem.
Initial platelet aggregation in the complex shear environment of a punctured vessel model C. J. Spieker, G. Závodszky, C. Mouriaux, P. H. Mangin, and A. G. Hoekstra, PLOS Computational Biology (2023) DOI: 10.1063/5.0157814 A combined in silico and in vitro platform is developed to study the hemostatic response to microneedle-induced vessel puncture. This platform integrates a cell-resolved blood flow model for detailed flow and cell distribution analysis with a novel punctured vessel flow chamber for experimental validation. Results from simulations and experiments reveal insights into platelet transport and aggregation dynamics in the wound area under various puncture conditions. Emphasis is placed on shear rate, elongational flow, and cell distributions, with discussions on the effects of wound size and pressure difference. Simulation findings indicate asymmetric cell distributions around the wound neck, confirmed by increased platelet aggregation at the distal side in experiments. This platform offers valuable insights into primary hemostasis and arterial thrombosis mechanisms.
Machine learning coarse-grained potentials of protein thermodynamics Majewski; Maciej; Pérez; Adrià; Thölke; Philipp; Doerr; Stefan; Charron; Nicholas E.; Giorgino; Toni; Husic; Brooke E.; Clementi; Cecilia; Noe; Frank; De Fabritiis; Gianni, Nature Communications (2023) DOI: 10.1038/s41467-023-41343-1 Understanding protein dynamics is a complex challenge crucial for deciphering structure-function relationships in biological processes. In this study, we address this challenge by employing artificial neural networks to construct coarse-grained molecular potentials grounded in statistical mechanics. Using an extensive dataset of unbiased all-atom molecular dynamics simulations across twelve proteins, we train these models, achieving significant acceleration of dynamics while maintaining system thermodynamics. Coarse-grained simulations successfully identify relevant structural states comparable to all-atom systems, and a single potential can encompass multiple proteins, even capturing experimental features of mutated proteins. These findings highlight the potential of machine learning-based coarse-grained potentials to simulate and elucidate protein dynamics.
Mesh U-Nets for 3D Cardiac Deformation Modeling Beetz, Marcel; Acero, Jorge Corral; Banerjee, Abhirup; Eitel, Ingo; Zacur, Ernesto; Lange, Torben; Stiermaier, Thomas; Evertz, Ruben; Backhaus, Sören J; Thiele, Holger; , International Workshop on Statistical Atlases and Computational Models of the Heart (2023) DOI: 10.1007/978-3-031-23443-9_23 During the cardiac cycle, intricate 3D deformations of the heart anatomy occur, offering insights into various cardiovascular pathologies, including myocardial infarction. While volume-based metrics like ejection fraction are common in clinical assessment, they offer limited insight into localized 3D changes. This study introduces a novel geometric deep learning method, mesh U-Net, which combines mesh-based convolution and pooling with U-Net-inspired architecture to analyze 3D ventricular shapes directly. Trained to model both contraction and relaxation, the network predicts cardiac anatomy at end-systole from end-diastole and vice versa. Evaluation on a multi-center cardiac MRI dataset reveals accurate predictions close to image resolution, with favorable performance compared to benchmarks, requiring fewer parameters and smaller data sizes. Additionally, the mesh U-Net captures subpopulation-specific differences in mechanical deformation patterns related to myocardial infarction types and clinical outcomes.
Multi-class point cloud completion networks for 3D cardiac anatomy reconstruction from cine magnetic resonance images Marcel Beetz; Abhirup Banerjee; Julius Ossenberg-Engels; Vicente Grau, Medical Image Analysis (2023) DOI: 10.1016/j.media.2023.102975 Cine magnetic resonance imaging (MRI) is the standard for cardiac anatomy and function assessment, but its limitation to 2D slices restricts comprehensive analysis. This paper introduces a novel automatic surface reconstruction pipeline for generating 3D cardiac anatomy meshes from raw cine MRI data. Central to this pipeline is a multi-class point cloud completion network (PCCN), which addresses sparsity and misalignment issues in a unified model. Evaluation on synthetic data demonstrates PCCN’s effectiveness, achieving reconstruction errors comparable to image resolution and outperforming benchmark methods. Applied to 1000 UK Biobank subjects, the pipeline produces accurate biventricular heart meshes with clinical metrics aligned with prior studies. Additionally, the approach exhibits robustness in handling common outlier conditions.
Multi-modality cardiac image computing: A survey Lei Li; Wangbin Ding; Liqin Huang; Xiahai Zhuang; Vicente Grau, Medical Image Analysis (2023) DOI: 10.1016/j.media.2023.102869 This paper provides a comprehensive review of multi-modality imaging in cardiology, emphasizing its role in enhancing diagnosis accuracy and improving patient management. Challenges include inter-modality misalignment and integration of information from different modalities. The review focuses on computing methodologies for registration, fusion, and segmentation tasks, highlighting their potential applications in clinical workflows such as guiding interventions and assessing myocardial viability. However, challenges remain, including missing modality data, modality selection, and uniform analysis of different modalities. Future research will address these issues and further integrate multi-modality imaging into clinical practice.
Multi-objective point cloud autoencoders for explainable myocardial infarction prediction Beetz, Marcel; Banerjee, Abhirup; Grau, Vicente, International Conference on Medical Image Computing and Computer-Assisted Intervention (2023) DOI: 10.48550/arxiv.2307.11017 In this study, we propose a novel approach, the multi-objective point cloud autoencoder, for predicting myocardial infarction (MI) based on 3D point cloud representations of cardiac anatomy and function. This method aims to address the limitations of existing image-based biomarkers by capturing more complex patterns in the heart’s 3D anatomy. The architecture of our approach includes multiple task-specific branches connected by a low-dimensional latent space, allowing for effective multi-objective learning of both reconstruction and MI prediction. By leveraging point cloud-based deep learning operations, our method enables efficient multi-scale feature learning directly on high-resolution anatomy point clouds.
NNP/MM: Accelerating Molecular Dynamics Simulations with Machine Learning Potentials and Molecular Mechanics Raimondas Galvelis; Alejandro Varela-Rial; Stefan Doerr; Roberto Fino; Peter Eastman; Thomas E. Markland; John D. Chodera; Gianni De Fabritiis, Journal of Chemical Information and Modeling (2023) DOI: 10.1021/acs.jcim.3c00773 We present an optimized implementation of the hybrid method (NNP/MM) to enhance the efficiency of biomolecular simulations. This method combines a neural network potential (NNP) with molecular mechanics (MM), effectively reducing computational costs while preserving accuracy. Specifically, NNP is employed to model a selected portion of the system, such as a small molecule, while MM is utilized for the remaining components. To demonstrate the efficacy of our NNP/MM implementation, we conducted molecular dynamics (MD) simulations on various protein–ligand complexes and metadynamics (MTD) simulations on a ligand. Our approach significantly accelerates simulation speed, achieving a fivefold improvement compared to conventional methods. Furthermore, we report the longest simulations to date for this class of studies, with a combined sampling of 1 μs for each complex. In summary, our optimized NNP/MM approach offers a promising solution to the computational challenges encountered in biomolecular simulations, facilitating more efficient and accurate investigations of protein-ligand interactions.
Orchestration of multiscale models for computational oncology Varella, Vinicius, Zenodo (2023) DOI: 10.5281/zenodo.8025198 PRIMAGE is a large EU project that aims to support personalized childhood cancer diagnosis and prognosis. The goal is to predict the growth of solid tumors using multiscale in-silico technologies. The project proposes an open cloud-based platform to support decision-making in the clinical management of pediatric cancers. The orchestration of predictive models is generally complex and would require a software framework that supports and facilitates such tasks. The present work proposes the development of an updated framework, referred to herein as VPH-HFv3, as a part of the PRIMAGE project. This framework, a complete rewriting with respect to the previous versions, aims to orchestrate several models, which are in concurrent development, using an architecture as simple as possible, easy to maintain, and with high reusability. This sort of problem generally requires unfeasible execution times. To overcome this problem, a strategy of particularization was developed, which maps the upper-scale model results into a smaller number, and homogenization which does the inverse way and analyzed the accuracy of this approach.
Point2Mesh-Net: Combining Point Cloud and Mesh-Based Deep Learning for Cardiac Shape Reconstruction Beetz, Marcel; Banerjee, Abhirup; Grau, Vicente; , International Workshop on Statistical Atlases and Computational Models of the Heart (2023) DOI: 10.1007/978-3-031-23443-9_26 Point2Mesh-Net, a novel geometric deep learning method, is introduced to efficiently and accurately transform 2D MRI slices into 3D cardiac surface meshes. Leveraging recent advancements in point cloud and mesh-based deep learning, the network directly processes sparse MRI contours and generates 3D triangular surface meshes suitable for follow-up tasks. Its hierarchical architecture enables multi-scale feature learning, overcoming challenges such as data sparsity and slice misalignment. Evaluation on synthetic and real MRI datasets demonstrates promising results, with surface distances between reconstructed and gold standard meshes below image resolution, validating its potential for clinical applications.
Post-Infarction Risk Prediction with Mesh Classification Networks Beetz, Marcel; Acero, Jorge Corral; Banerjee, Abhirup; Eitel, Ingo; Zacur, Ernesto; Lange, Torben; Stiermaier, Thomas; Evertz, Ruben; Backhaus, Sören J; Thiele, Holger; , International Workshop on Statistical Atlases and Computational Models of the Heart (2023) DOI: 10.1007/978-3-031-23443-9_27 Post-myocardial infarction (MI) patients face the risk of major adverse cardiac events (MACE), typically assessed through global image-based biomarkers like ejection fraction. However, such metrics overlook subtle, localized shape disparities in 3D cardiac anatomy, affecting predictive accuracy. This study introduces a novel geometric deep learning method tailored for high-resolution 3D cardiac anatomy meshes to forecast MACE within a year post-infarction. Employing a hierarchical, multi-scale structure, the model efficiently processes surface mesh data, enabling comprehensive local and global feature learning. Evaluation on a multi-center dataset reveals superior performance compared to clinical benchmarks, with notable enhancements in area under the receiver operating characteristic (AUROC) curve, showcasing promising predictive capabilities amidst class imbalances.
Simulating Initial Steps of Platelet Aggregate Formation in a Cellular Blood Flow Environment C. J. Spieker, K. Asteriou, and G. Zavodszky, Lecture Notes in Computer Science (2023) DOI: 10.1007/978-3-031-36024-4_26 This study introduces a cell-based model to investigate the initial steps of clot formation, crucial for understanding hemostasis and thrombosis. Integrated into the HemoCell blood flow model, it focuses on platelet adhesion and aggregation under arterial flow conditions. Unlike existing models, it emphasizes the mechanical environment’s role in aggregate formation, utilizing simplified constraint-dependent platelet binding. Notably, it accounts for factors like elongational flows and platelet margination, enabled by the cell-resolved scale. Detailed implementation and characteristic behavior of the model at various threshold values are discussed, offering insights into clot formation mechanisms and their hemodynamic influences.
Teaching Medical Students to Use Supercomputers: A Personal Reflection. Townsend-Nicholson A, Methods in Molecular Biology (2023) DOI: 10.1007/978-1-0716-3449-3_20 The inception of the CompBioMed project in 2016 sparked the development of the CompBioMed Education and Training Programme, aimed at integrating supercomputing education into medical and biomedical curricula. This holistic initiative, fostered within the CompBiomed Centre of Excellence, engages experts from various disciplines including experimental researchers, computer scientists, clinicians, HPC centers, and industrial partners. Refinements to the program, detailed in this chapter, reflect six years of highs and lows in its delivery. Suggestions for overcoming barriers include feasible measures to bridge the gap for users unfamiliar with high-performance computing, thus advancing computational literacy in medical and biomedical fields.
Unified directional parabolic-accurate lattice Boltzmann boundary schemes for grid-rotated narrow gaps and curved walls in creeping and inertial fluid flows Irina Ginzburg; Goncalo Silva; Francesco Marson; Bastien Chopard; Jonas Latt, Physical Review E (2023) DOI: 10.1103/physreve.107.025303 This work aims to enhance lattice Boltzmann Dirichlet velocity boundary schemes for improved accuracy, stability, and mass conservation in various fluid flow scenarios, including straight walls, curved surfaces, and narrow gaps. Two boundary classes, LI+ and EMR, are introduced to address these challenges. LI+ incorporates directional rules and nonequilibrium local corrections for different flow regimes, while EMR extends the accuracy to grid-rotated linear Couette and Poiseuille flows. The methodology ensures improved accuracy and stability without compromising computational efficiency, demonstrated through numerical tests across different flow configurations. Overall, these advancements offer robust solutions for both creeping and inertial flow regimes.
What determines the optimal pharmacological treatment of atrial fibrillation? Insights from in silico trials in 800 virtual atria Albert Dasí, Michael T.B. Pope, Rohan S. Wijesurendra, Tim R. Betts, Rafael Sachetto, Alfonso Bueno-Orovio and Blanca Rodriguez, Journal of Physiology (2023) DOI: 10.1113/JP284730 This study leverages simulations of atrial fibrillation (AF) in a virtual cohort of 800 atria to pinpoint patient characteristics guiding the selection of anti-arrhythmic drugs. Variability in electrophysiology and low voltage areas (LVA) was considered, with validation against experimental and clinical data. AF sustained in 62% of atria, with large inward rectifier K+ current (IK1) and Na+/K+ pump (INaK) densities linked to AF maintenance. In severely remodelled left atria with extensive LVA, higher IK1 was required, influencing rotor localization. Atrial refractoriness, modulated by ICaL and INa, determined treatment success. Results suggest inward currents as crucial for patient stratification and LVA extension for accurate AF phenotyping.
Long timescale ensemble methods in molecular dynamics: Ligand-protein interactions and allostery in SARS-CoV-2 targets A. Bhati, A. Hoti, A. Potterton, M. K. Bieniek, P. V. Coveney, Journal of Chemical Theory and Computation (2023) DOI: 10.1021/acs.jctc.3c00020 We subject a series of five protein-ligand systems which contain important SARS- CoV-2 targets – 3-chymotrypsin-like protease, papain-like protease and adenosine ribose phosphatase – to long-timescale and adaptive sampling molecular dynamics simulations. By performing ensembles of ten or twelve 10-microsecond simulations for each system, we accurately and reproducibly determine ligand binding sites, both crystallographically resolved and otherwise, thereby discovering binding sites that can be exploited for drug discovery. We also report robust, ensemble-based observation of conformational changes that occur at the main binding site of 3CLPro due to the presence of another ligand at an allosteric binding site explaining the underlying cascade of events responsible for its inhibitory effect. Using our simulations, we have discovered a novel allosteric mechanism of inhibition for a ligand known to bind only at the substrate binding site. Due to the chaotic nature of molecular dynamics trajectories, individual trajectories do not allow for accurate or reproducible elucidation of macroscopic expectation values. Unprecedentedly at this timescale, we compare the statistical distribution of protein-ligand contact frequencies for these ten/twelve 10-microsecond trajectories and find that over 90% of trajectories have significantly different contact frequency distributions. Furthermore, using a direct binding free energy calculation protocol, we determine the ligand binding free energies for each of the identified sites using long-timescale simulations. The free energies differ by 0.77 to 7.26 kcal/mol across individual trajectories depending on the binding site and the system. We show that although this is the standard way such quantities are currently reported at long-timescale, individual simulations do not yield reliable free energies. Ensembles of independent trajectories are necessary to overcome the aleatoric uncertainty in order to obtain statistically meaningful and reproducible results. Our findings here are generally applicable to all molecular dynamics based applications and not confined to the free energy methods used in this study. Finally, we compare the application of different free energy methods to these systems and discuss their advantages and disadvantages.
Structure and dynamics of an archetypal DNA nanoarchitecture revealed via cryo-EM and molecular dynamics simulations K. Ahmad, A. Javed, C. Lanphere, P. V. Coveney, E. V. Orlova, S. Howorka, Nature Communications (2023), DOI: 10.1038/s41467-023-38681-5 DNA can be folded into rationally designed, unique, and functional materials. To fully realise the potential of these DNA materials, a fundamental understanding of their structure and dynamics is necessary, both in simple solvents as well as more complex and diverse anisotropic environments. Here we analyse an archetypal six-duplex DNA nanoarchitecture with single-particle cryo-electron microscopy and molecular dynamics simulations in solvents of tunable ionic strength and within the anisotropic environment of biological membranes. Outside lipid bilayers, the six-duplex bundle lacks the designed symmetrical barrel-type architecture. Rather, duplexes are arranged in non-hexagonal fashion and are disorted to form a wider, less elongated structure. Insertion into lipid membranes, however, restores the anticipated barrel shape due to lateral duplex compression by the bilayer. The salt concentration has a drastic impact on the stability of the inserted barrel-shaped DNA nanopore given the tunable electrostatic repulsion between the negatively charged duplexes. By synergistically combining experiments and simulations, we increase fundamental understanding into the environment-dependent structural dynamics of a widely used nanoarchitecture. This insight will pave the way for future engineering and biosensing applications.
A multiscale computational framework to evaluate flow alterations during mechanical thrombectomy for treatment of ischaemic stroke I. Benemerito, A. Mustafa, N. Wang, A. P. Narata, A. Narracott, Marzo Alberto, Frontiers in Cardiovascular Medicine (2023), DOI:10.3389/fcvm.2023.1117449
The treatment of ischaemic stroke increasingly relies upon endovascular procedures known as mechanical thrombectomy (MT), which consists in capturing and removing the clot with a catheter-guided stent while at the same time applying external aspiration with the aim of reducing haemodynamic loads during retrieval. In this study we present a multiscale computational framework to simulate MT procedures. The developed framework can provide quantitative assessment of clinically relevant quantities such as flow in the retrieval path and can be used to find the optimal procedural parameters that are most likely to result in a favorable clinical outcome.
Statistical Properties of a Virtual Cohort for In Silico Trials Generated with a Statistical Anatomy Atlas A. Mattina, F. Baruffaldi, M. Taylor & M. Viceconti, Ann Biomed Eng (2023), DOI:10.1007/s10439-022-03050-8
Osteoporosis-related hip fragility fractures are a catastrophic event for patient lives but are not frequently observed in prospective studies, and therefore phase III clinical trials using fractures as primary clinical endpoint require thousands of patients enrolled for several years to reach statistical significance. A novel answer to the large number of subjects needed to reach the desired evidence level is offered by In Silico Trials, that is, the simulation of a clinical trial on a large cohort of virtual patients, monitoring the biomarkers of interest. In this work we investigated if statistical aliasing from a custom anatomy atlas could be used to expand the patient cohort while retaining the original biomechanical characteristics.
TIES 2.0: a Dual-Topology Open Source Relative Binding Free Energy Builder with Web Portal M. Bieniek, A. Wade, A. Bhati, S. Wan and P. V. Coveney, Journal of Chemical Information and Modeling (2023) DOI:10.1021/acs.jcim.2c01596 Relative binding free energy (RBFE) calculations are widely used to aid the process of drug discovery. TIES, Thermodynamic Integration with Enhanced Sampling, is a dual-topology approach to RBFE calculations with support for NAMD and OpenMM molecular dynamics engines. The software has been thoroughly validated on publicly available datasets. Here we describe the open source software along with a web portal (https://ccs-ties.org) that enables users to perform such calculations correctly and rapidly.

2022

Title Citation Summary
Personalized pathology test for Cardio-vascular disease: Approximate Bayesian computation with discriminative summary statistics learning Dutta R, Zouaoui Boudjeltia K, Kotsalos C, Rousseau A, Ribeiro de Sousa D, et al., PLOS Computational Biology (2022) DOI: 10.1371/journal.pcbi.1009910 Cardio/cerebrovascular diseases (CVD) have become one of the major health issue in our societies. But recent studies show that the present pathology tests to detect CVD are ineffectual as they do not consider different stages of platelet activation or the molecular dynamics involved in platelet interactions and are incapable to consider inter-individual variability. Here we propose a stochastic platelet deposition model and an inferential scheme to estimate the biologically meaningful model parameters using approximate Bayesian computation with a summary statistic that maximally discriminates between different types of patients.
Dynamic resource allocation for efficient parallel CFD simulations G. Houzeaux; R.M. Badia; R. Borrell; D. Dosimont; J. Ejarque; M. Garcia-Gasulla; V. López, Computers & Fluids (2022) DOI: 10.1016/j.compfluid.2022.105577 Supercomputer users employing Computational Fluid Dynamics (CFD) often face challenges in selecting the optimal number of subdomains (partitions) for MPI-based parallelization. Typically, users resort to rule-of-thumb methods, such as setting a minimum number of elements or cells per subdomain to maintain parallel efficiency above a certain threshold, like 80%. However, this approach can be subjective and may lead to resource wastage, especially when users lack knowledge about the best practices for a specific code. This study introduces an elastic computing methodology that dynamically adjusts the allocated resources for a simulation at runtime. The methodology relies on a runtime measure of communication efficiency during execution. Analytical estimates are used to expand or reduce resources to ensure efficient simulation execution. By adapting resource allocation based on communication efficiency, the proposed approach aims to optimize simulation performance without the need for manual intervention or predefined thresholds.
Hybrid parallelization of molecular dynamics simulations to reduce load imbalance J. Morillo, M. Vassaux, P. V. Coveney, M. Garcia-Gasulla, Journal of Supercomputing (2022) DOI: 10.1007/s11227-021-04214-4 The predominant method for enabling parallel simulations in molecular dynamics involves spatial domain decomposition, dividing the physical geometry into boxes, typically one per processor. However, this technique can lead to computational load imbalances when particle distribution or computational costs per particle are non-uniform. The study explores the advantages of employing a hybrid MPI+OpenMP model to address this load imbalance issue. Focusing on LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator), a leading molecular dynamics simulator, the research extends its existing OpenMP implementation and optimizes it. Three setups are evaluated: MPI-only, MPI with LAMMPS’ built-in balance mechanism, and a hybrid setup utilizing the enhanced OpenMP version. Performance comparison includes five standard benchmarks from LAMMPS distribution and two additional test cases. Results indicate that the hybrid approach effectively mitigates load balancing problems, showcasing a 50% improvement over MPI-only in highly imbalanced scenarios. In contrast, the LAMMPS balance mechanism shows a 43% improvement. Moreover, the hybrid setup enhances simulations encountering challenges beyond load imbalance.
A Multi-view Crossover Attention U-Net Cascade with Fourier Domain Adaptation for Multi-domain Cardiac MRI Segmentation Marcel Beetz; Jorge Corral Acero; Vicente Grau, Lecture Notes in Computer Science (2022) DOI: 10.1007/978-3-030-93722-5_35 Cardiac image segmentation is pivotal for assessing cardiac morphology and quantifying image-based biomarkers in clinical practice. While deep learning methods show promising performance on single-domain cine MRI datasets, their accuracy diminishes in complex multi-domain settings, limiting clinical utility. To address this, we propose a novel multi-view crossover cascade approach integrating shape and appearance augmentations for effective multi-domain cardiac image segmentation. Our cascade includes two Attention U-Net paths sharing information across views and an intermediate heart location crop for variance reduction and label balance improvement. We introduce various shape augmentations and histogram matching, along with multi-scale Fourier Domain Adaptation for cardiac image analysis. Evaluation on the M&Ms-2 challenge cine MRI dataset demonstrates improved performance over a U-Net benchmark, with respective Dice score increases of 0.02 and 0.03.
Automated 3D Whole-Heart Mesh Reconstruction From 2D Cine MR Slices Using Statistical Shape Model Banerjee, A; Zacur, E; Choudhury, RP; Grau, V, 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (2022) DOI: 10.1109/embc48229.2022.9871327 Cardiac magnetic resonance (CMR) imaging is a key modality for diagnosing cardiovascular diseases, often producing high-resolution 2D images of heart tissues in separated planes. However, this cine protocol may not adequately capture the entire heart, particularly both atria. This paper introduces a novel approach for automated patient-specific 3D whole-heart mesh reconstruction from a limited number of 2D cine CMR slices, leveraging a statistical shape model (SSM). After extracting heart contours from 2D cine slices, the SSM is fitted over the sparse heart contours in 3D space, providing an initial 3D whole-heart mesh representation, which is then deformed to minimize distance from the contours, yielding the final reconstructed mesh. Evaluation on a cohort of 30 subjects from the UK Biobank study demonstrates high-quality 3D whole-heart meshes, with average contours-to-surface distance below image resolution and clinical metrics within acceptable ranges. This automated reconstruction has significant clinical relevance, facilitating cardiac diagnosis, treatment planning, virtual surgery, and biomedical simulations.
“Automated Torso Contour Extraction from Clinical Cardiac MR Slices for 3D Torso Reconstruction” Hannah J. Smith, Abhirup Banerjee, Robin P. Choudhury and Vicente Grau, 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (2022) DOI: 10.1109/EMBC48229.2022.9871643 The electrocardiogram (ECG) is pivotal for diagnosing cardiac electrical abnormalities, but its interpretation is influenced by torso geometry variations. We propose a novel pipeline that utilizes standard cardiac magnetic resonance images to reconstruct the torso and heart, enabling ECG recreation considering both torso and cardiac anatomy. This involves automated extraction of torso contours, achieved through a method combining an initial u-net segmenter with a refined contour extraction network. Evaluation on cross-validation and independent test datasets demonstrates the efficacy of the approach, particularly with the refinement network utilizing two-channel input, improving performance significantly on the independent test set. This method holds clinical relevance by enabling ECG simulations with personalized torso geometry, known to impact QRS parameters significantly. It offers potential for developing a clinical tool that incorporates torso geometry in ECG interpretation, enhancing diagnostic accuracy and personalized patient care.
Combined Generation of Electrocardiogram and Cardiac Anatomy Models Using Multi-Modal Variational Autoencoders Beetz, M; Banerjee, A; Sang, Y; Grau, V, IEEE 19th International Symposium on Biomedical Imaging (ISBI) (2022) DOI: 10.1109/isbi52829.2022.9761590 This work introduces a novel multi-modal variational autoencoder (VAE) capable of processing combined physiology and bitemporal anatomy information, specifically electrocardiograms (ECG) and 3D biventricular point clouds. The method achieves high reconstruction accuracy on a UK Biobank dataset, with Chamfer distances between predicted and input anatomies below the underlying image resolution, and ECG reconstructions outperforming a state-of-the-art benchmark approach specialized in ECG generation. Generative ability evaluation shows comparable populations of generated and gold standard anatomies, ECGs, and combined anatomy-ECG data in terms of common clinical metrics and maximum mean discrepancies.
Directional lattice Boltzmann boundary conditions Marson, Francesco, Thesis/Dissertation (2022) DOI: 10.13097/archive-ouverte/unige:160770 The thesis aims to propose highly accurate curved boundary conditions, reducing the need for grid refinement while ensuring compatibility with massively parallel CPU or GPU systems. Novel boundary conditions, called Enhanced Local Interpolation (ELI), result from an extensive analysis and improvement of lattice Boltzmann method directional boundary conditions (DBC). ELI, along with additional corrections to DBC, enhance accuracy, particularly in low Reynolds number scenarios. These adjustments are algorithmically and physically local, requiring no information recovery outside the boundary node and allowing modeling of narrow gaps with a single computational node.
Effect of particularisation size on the accuracy and efficiency of a multiscale tumours’ growth model Vinicius Varella; Barbara de M. Quintela; Marek Kasztelnik; Marco Viceconti, International Journal for Numerical Methods in Biomedical Engineering (2022) DOI: 10.1002/cnm.3657 In silico medicine models often face challenges with long execution times due to coupling multiple space-time scales, leading to computational inefficiencies. To address this, a common practice involves using particularization and homogenization operators to simplify the coupling between scales. This study aims to find the simplest scheme to couple a whole solid tumor growth model with a cell-tissue biology model while minimizing approximation error and computational cost. By applying a basic binning strategy to an initial dataset resembling real neuroblastomas, we found that even this simple approach significantly reduces computational cost with minimal approximation errors.
Experimental validation of a subject-specific finite element model of lumbar spine segment using digital image correlation Garavelli C, Curreli C, Palanca M, Aldieri A, Cristofolini L, Viceconti M, PLOS ONE (2022) DOI: 10.1371/journal.pone.0272529 This study outlines a procedure for creating patient-specific finite element models to predict fracture risk in vertebral bone affected by pathologies like cancer metastasis. Using quantitative computed tomography images of a cadaveric lumbar spine segment with vertebral metastatic lesions, the model’s kinematics were validated against experimental data. Boundary conditions mimicked compression-flexion loading. Results showed good agreement between experimental and simulated vertebral surface displacements (R2 > 0.9, RMSE% < 8%), demonstrating the model’s potential in predicting spine segment displacement under physiological conditions. This marks a crucial step in assessing the credibility of clinical decision-support models.
Generating Subpopulation-Specific Biventricular Anatomy Models Using Conditional Point Cloud Variational Autoencoders Marcel Beetz; Abhirup Banerjee; Vicente Grau, Lecture Notes in Computer Science (2022) DOI: 10.1007/978-3-030-93722-5_9 Introducing a novel geometric deep learning approach within the variational autoencoder (VAE) framework, this study focuses on accurately encoding, reconstructing, and synthesizing 3D surface models of biventricular anatomy. Operating directly on memory-efficient point clouds, the method can process multiple cardiac substructures simultaneously in a multi-class setting. Additionally, it incorporates subpopulation-specific characteristics as conditional inputs, enabling the generation of personalized anatomies. Evaluation on a UK Biobank-derived dataset demonstrates high reconstruction quality, with average Chamfer distances below image pixel resolution. The method successfully synthesizes virtual populations of realistic hearts aligned with clinical norms and explores variations in the latent space, revealing interpretable changes in cardiac shapes and sizes.
Generation of 12-lead electrocardiogram with subject-specific, image-derived characteristics using a conditional variational autoencoder Sang, Yuling; Beetz, Marcel; Grau, Vicente; , IEEE 19th International Symposium on Biomedical Imaging (ISBI) (2022) DOI: 10.1109/ISBI52829.2022.9761431 This paper introduces a conditional variational autoencoder (cVAE) for automatically generating realistic 12-lead ECG signals. Distinguishing itself from prior work, our method generates complete 12-lead studies and allows adjustment of generated ECGs based on specific subject characteristics, including age, sex, and Body Mass Index (BMI). Notably, our model incorporates imaging information by integrating heart position and orientation as input conditions, enabling analysis of anatomical influences on ECG morphology. Demonstrating high accuracy and sensitivity to different conditions, our model extracts a ten-dimensional latent space containing interpretable features of the 12 ECG leads.
Gordon Research Conference on Cardiac Arrhythmia Mechanisms 2023: early career investigators’ views on emerging concepts and technologies Ahmed Ramadan, Enaam Chleilat, Hector Martinez-Navarro, Jaclyn Brennan-McLean, Jaël Copier, Jessica Caldwell, Joachim Greiner, Patricia Martínez Díaz, Vladimír Sobota, Journal of Physiology (2022) DOI: 10.1113/JP284666 Personalized medicine, focusing on patient-specific diagnosis and treatment, has gained momentum, considering factors like sex, age, comorbidities, and genetic profiles. Historically, research predominantly on males led to biases in medicine, overlooking potential sex-based differences in physiology. In arrhythmia research, the lack of female representation hampers understanding and may lead to suboptimal diagnoses and treatments. Studies highlight sex-specific targets for arrhythmia treatments, while large-scale genomic datasets provide avenues to explore sex’s role in cardiovascular disease. Considering situations where sex hormones are altered, like with age, is crucial, as they can impact cardio-protection against arrhythmias. Continued research into personalized medicine is imperative.
Hotspot Identification and Drug Design of Protein–Protein Interaction Modulators Using the Fragment Molecular Orbital Method Monteleone S, Fedorov DG, Townsend-Nicholson A, Southey M, Bodkin M, Heifetz A, Journal of Chemical Information and Modeling (2022) DOI: 10.1021/acs.jcim.2c00457 Protein-protein interactions (PPIs) are vital for protein function and can be targeted for drug discovery. Despite a vast number of known PPIs, few PPI modulators have reached clinical use. Fragment molecular orbital (FMO) quantum mechanics offers a cost-effective method to assess the strength and nature of molecular interactions at the protein-protein interface. Integrating FMO with PPI exploration (FMO-PPI) helps identify critical binding residues (hotspots). Validation with protein complexes aligns FMO-PPI results with mutagenesis data, revealing three types of critical interactions: intermolecular, intramolecular, and water-mediated. FMO-PPI findings support structure-based drug design of PPI modulators (SBDD-PPI).
In-silico drug trials for precision medicine in atrial fibrillation: From ionic mechanisms to electrocardiogram-based predictions in structurally-healthy human atria Albert Dasí, Aditi Roy, Rafael Sachetto, Julia Camps, Alfonso Bueno-Orovio and Blanca Rodriguez, Frontiers in Physiology (2022) DOI: 10.3389/fphys.2022.966046 In this study, in-silico drug trials were conducted using a population of structurally-healthy human atria models to explore AF inducibility, sustainability, and response to pharmacological treatment. Results revealed that AF episodes were induced and sustained by specific ionic current properties, particularly those related to action potential duration (APD) restitution and excitability. Decreasing excitability through prolonged refractoriness facilitated pharmacological cardioversion. In-silico trials identified optimal treatments for individual atrial electrophysiological properties and showed >70% accuracy in predicting cardioversion outcomes using either ionic current profiles or ECG metrics. These findings suggest that AF inducibility and response to treatment can be predicted based on atrial ionic properties, highlighting the potential for personalized therapeutic approaches.
Inference of Number and Location of Purkinje Root Nodes and Ventricular Conduction Properties from Clinical 12-Lead ECGs for Cardiac Digital Twinning Camps, Julia; Wang, Zhinuo Jenny; Sebastian, Rafael; Zhou, Xin; Lawson, Brodie; Berg, Lucas Arantes; Burrage, Kevin; Grau, Vicente; Weber, Rodrigo; Rodriguez, Blanca; , Computing in Cardiology (CinC) (2022) DOI: 10.22489/CinC.2022.235 This study introduces novel computational techniques to infer physiological Purkinje network properties from clinical CMR and ECG data. Using Eikonal simulations and an extended sequential Monte Carlo approximate Bayesian computation-based inference method, the study achieves an improved match in simulated QRS complexes to clinical 12-lead ECGs, with Pearson’s correlation coefficients of 0.89. These findings pave the way for the development of cardiac digital twins that incorporate Purkinje network information, enabling enhanced risk stratification and clinical decision-making.
Interpretable cardiac anatomy modeling using variational mesh autoencoders Beetz M, Corral Acero J, Banerjee A, Eitel I, Zacur E, Lange T, Stiermaier T, Evertz R, Backhaus SJ, Thiele H, Bueno-Orovio A, Lamata P, Schuster A, Grau V, Frontiers in Cardiovascular Medicine (2022) DOI: 10.3389/fcvm.2022.983868 In this study, we introduce a novel approach called the variational mesh autoencoder (mesh VAE) for modeling population-wide variations in cardiac shapes. This method incorporates multi-scale graph convolutions and mesh pooling layers into a hierarchical VAE framework, enabling efficient processing of surface mesh representations of cardiac anatomy. The mesh VAE demonstrates strong performance in reconstructing 3D cardiac meshes from a dataset of over 1,000 patients with acute myocardial infarction, with low mean surface distances between input and reconstructed meshes. Compared to a voxelgrid-based deep learning benchmark, the mesh VAE achieves superior reconstruction accuracy while requiring less memory. Additionally, we explore the interpretability of the mesh VAE’s latent space and its ability to improve the prediction of major adverse cardiac events compared to clinical benchmarks. Finally, we demonstrate the method’s capability to generate realistic virtual populations of cardiac anatomies, showcasing good alignment with gold standard mesh populations across multiple clinical metrics.
Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology Beetz, M; Banerjee, A; Grau, V, Frontiers in Physiology (2022) DOI: 10.3389/fphys.2022.886723 This work introduces a novel multi-domain Variational Autoencoder (VAE) network to capture combined Electrocardiogram (ECG) and Magnetic Resonance Imaging (MRI)-based 3D anatomy information. Specifically designed branches address challenges in each input domain, enabling efficient encoding, reconstruction, and synthesis of multi-domain cardiac signals. Evaluation on a United Kingdom Biobank dataset demonstrates high reconstruction accuracy, outperforming single-domain benchmarks significantly. The VAE generates realistic virtual populations with good alignment in clinical metrics and interpretable latent space. Furthermore, combining anatomy and ECG representation improves cardiac disease classification performance by 3.9% in AUROC curve over single-domain approaches.
Personalised 3D Assessment of Trochanteric Soft Tissues Improves Hip Fracture Classification Accuracy A Aldieri, M Terzini, AL Audenino, C Bignardi, M Paggiosi, R Eastell, M Viceconti, P Bhattacharya, Annals of Biomedical Engineering (2022) DOI: 10.1007/s10439-022-02924-1 This study explores the impact of personalized and orientation-specific Soft Tissue Thickness (STT) estimates, derived from CT scans, on fracture classification accuracy using the absolute risk of hip fracture (ARF0). Results indicate that personalized 3D STT estimates consistently exceed BMI-based estimates. Fracture classification accuracy improves when personalized 3D STT estimates are used, with an AUC of 0.87 compared to 0.85 using BMI-based estimates. Incorporating orientation-specificity of CT-based STT further enhances classification accuracy, yielding an AUC of 0.86. Overall, the study underscores the importance of individualized assessments in predicting hip fracture risk.
PlayMolecule Glimpse: Understanding Protein–Ligand Property Predictions with Interpretable Neural Networks Alejandro Varela-Rial; Iain Maryanow; Maciej Majewski; Stefan Doerr; Nikolai Schapin; José Jiménez-Luna; Gianni De Fabritiis, Journal of Chemical Information and Modeling (2022) DOI: 10.1021/acs.jcim.1c00691 This study introduces an approach to visualize the contribution of individual input atoms to predictions made by convolutional neural networks (CNNs) in structure-based protein-ligand affinity prediction. Using KDEEP, a CNN model for binding affinity prediction, the method aims to enhance the interpretability of model predictions. Results indicate that KDEEP learns meaningful chemistry signals from the data, while also highlighting areas for model optimization. This approach provides valuable insights into the inner workings of CNN-based protein-ligand affinity prediction models.
Predicting 3D Cardiac Deformations with Point Cloud Autoencoders Marcel Beetz; Julius Ossenberg-Engels; Abhirup Banerjee; Vicente Grau, Lecture Notes in Computer Science (2022) DOI: 10.1007/978-3-030-93722-5_24 Heart mechanical contraction and relaxation are crucial for assessing cardiac health, yet their 3D patterns pose challenges for accurate evaluation, particularly on 2D images. This study introduces a novel geometric deep learning method to capture 3D biventricular deformations during end-diastolic (ED) and end-systolic (ES) phases. Utilizing an encoder-decoder structure with lightweight point cloud data, the model predicts ED outputs from ES inputs and vice versa. Validation on the UK Biobank cohort demonstrates promising performance with an average Chamfer distance of 1.66 ± 0.62 mm. Clinical metrics derived from predictions closely match gold standards, highlighting the method’s potential for assessing cardiac function across diverse populations.
Predicting antibiotic resistance in complex protein targets using alchemical free energy methods Brankin AE, Fowler PW, Journal of Computational Chemistry (2022) DOI: 10.1002/jcc.26979 Drug-resistant Mycobacterium tuberculosis, often arising from mutations in antibiotic target genes, presents a significant challenge in tuberculosis treatment. This study explores the use of relative binding free energy (RBFE) calculations to predict the impact of mutations on rifampicin and moxifloxacin susceptibility in M. tuberculosis RNA polymerase and DNA gyrase, respectively. While RBFE calculations show promise in predicting resistance for mutations causing large changes in binding free energy, their efficacy diminishes for subtle energy changes or mutations involving charged amino acids. Strategies to mitigate calculation errors in these scenarios are investigated, highlighting the method’s potential and limitations in predicting antibiotic resistance.
Proximal femur bone mineral density in osteoporotic patients a review of placebo groups in clinical trials Oliviero, Sara, Zenodo (2022) DOI: 10.5281/zenodo.7248963 This study conducts a systematic review of Bone Mineral Density (BMD) loss observed in placebo groups of clinical trials focusing on osteoporosis drug treatments. By analyzing this data, the study aims to characterize bone loss patterns in populations commonly enrolled in interventional trials. Additionally, the findings will inform simulations of disease progression in In Silico trials, providing valuable insights into the natural course of osteoporosis and aiding in the development and evaluation of potential treatment interventions.
Reconstructing 3D Cardiac Anatomies from Misaligned Multi-View Magnetic Resonance Images with Mesh Deformation U-Nets Beetz, Marcel; Banerjee, Abhirup; Grau, Vicente; , Proceedings of Machine Learning Research (2022) DOI: TBC This work introduces a novel method, the Mesh Deformation U-Net, for reconstructing high-quality 3D cardiac surface meshes from 2D MRI slices. By leveraging spectral graph convolutions and mesh sampling operations in a hierarchical encoder-decoder structure, this approach enables efficient multi-scale feature learning directly on mesh data. A targeted preprocessing step fits a template mesh to sparse MRI contours, while the Mesh Deformation U-Net corrects motion-induced slice misalignment using information from multiple MRI views and anatomical shape priors. Evaluation on synthetic datasets and cine MRI from the UK Biobank demonstrates superior performance in reconstructing cardiac anatomy, with small errors and realistic results.
Shear induced diffusion of platelets revisited Christos Kotsalos; Franck Raynaud; Jonas Lätt; Ritabrata Dutta; Frank Dubois; Karim Zouaoui Boudjeltia; Bastien Chopard, Frontiers in Physiology (2022) DOI: 10.3389/fphys.2022.985905 The traditional assumption of platelet transport in blood, following an advection-diffusion equation with a diffusion constant from the Zydney-Colton theory, is revisited in light of experimental findings and numerical simulations. By considering a fully resolved suspension of red blood cells and platelets under shear, we observe a non-trivial velocity distribution perpendicular to the flow, characterized by an exponential decrease in bulk followed by a power law tail. This distribution leads to platelet diffusion significantly higher than predicted by the Zydney-Colton theory, as confirmed by stochastic modeling. These findings challenge existing notions of platelet transport and highlight the need for refined models in understanding blood flow dynamics.
Structure based virtual screening: Fast and slow A Varela‐Rial, M Majewski, G De Fabritiis, Wiley Interdisciplinary Reviews (2022) DOI: 10.1002/wcms.1544 Virtual screening methods, notably molecular docking, have long been indispensable in early drug discovery, offering cost-effective solutions. While providing rapid insights from structural data, they often rely on assumptions, compromising accuracy. Conversely, physical-based molecular simulations offer precision but demand more resources. Both approaches address distinct facets of the same challenge. This review evaluates their capabilities, contexts, and limitations, citing various examples. It underscores the complementary roles of fast screening methods and slower, more detailed simulations in drug discovery endeavors.
Teaching genomics to life science undergraduates using cloud computing platforms with open datasets Poolman TM, Townsend-Nicholson A, Cain A, Biochemistry and Molecular Biology Education (2022) DOI: 10.1002/bmb.21646 The COVID-19 pandemic posed challenges for laboratory-based research projects in biochemistry undergraduate programs. To provide research opportunities, we engaged 80 students with limited computing experience in computational metagenomics projects using open datasets. Leveraging Google Colaboratory (Colab), a virtual computing environment, students accessed raw sequencing data and utilized QIIME2 for analysis and visualization. Colab’s user-friendly interface facilitated setup, synchronous sessions, and data sharing. We integrated Google Cloud Compute for projects requiring extensive computational resources. Moving forward, reanalyzing public data will be a standard component of research projects, enhancing students’ data science skills across various programming languages within the Colab environment.
The effect of stiffened diabetic red blood cells on wall shear stress in a reconstructed 3D microaneurysm Czaja, B., de Bouter, J., Heisler, M., Závodszky, G., Karst, S., Sarunic, M., Maberley, D., Hoekstra, A., Computer Methods in Biomechanics and Biomedical Engineering (2022) DOI: 10.1080/10255842.2022.2034794 This research employs cell-resolved blood flow simulations to investigate the impact of pulsatile flow on diabetic retinopathy progression, focusing on retinal microaneurysms. Using adaptive optics optical coherence tomography, a sidewall microaneurysm is segmented from volumetric data. Varying the microaneurysm neck width reveals its influence on red blood cell and platelet transport. Increasing red blood cell membrane stiffness simulates diabetic conditions, affecting flow dynamics. Calculations of wall shear stress and gradients highlight the impact of red blood cell stiffness, with increased values observed. Stiffer red blood cells penetrate the aneurysm more, altering platelet margination and penetration, potentially elucidating diabetic retinopathy mechanisms.
Understanding and improving risk assessment after myocardial infarction using automated left ventricular shape analysis Corral Acero, Jorge; Schuster, Andreas; Zacur, Ernesto; Lange, Torben; Stiermaier, Thomas; Backhaus, Sören J; Thiele, Holger; Bueno-Orovio, Alfonso; Lamata, Pablo; Eitel, Ingo; , Cardiovascular Imaging (2022) DOI: 10.1016/j.jcmg.2021.11.027 Left ventricular ejection fraction (LVEF) and end-systolic volume (ESV) are standard imaging biomarkers for post-acute myocardial infarction (AMI) risk assessment, yet they lack sensitivity to regional abnormalities. This study introduces novel 3-dimensional (3D) imaging descriptors of end-systolic (ES) shape and contraction for improved risk stratification in AMI survivors. Analyzing a multicenter cohort (n = 1,021), 3D shape markers outperformed ESV, while 3D contraction surpassed LVEF in predicting major adverse cardiac events (MACE). These descriptors enhanced risk prediction beyond traditional markers and identified specific impairments post-AMI, offering insights into shape and contraction patterns associated with MACE occurrence.
Reducing the Complexity of Musculoskeletal Models Using Gaussian Process Emulators I. Benemerito, E. Montefiori, A. Marzo, C. Mazzà, Applied Sciences DOI:10.3390/app122412932
Musculoskeletal models (MSKMs) are used to estimate the muscle and joint forces involved in human locomotion, often associated with the onset of degenerative musculoskeletal pathologies. In this work we have developed a methodology relying on Sobol’s sensitivity analysis (SSA) for ranking muscles based on their importance to the determination of the joint contact forces (JCFs) in a cohort of older women. Results show that there is a pool of muscles whose personalisation has little effects on the predictions of JCFs, allowing for a reduced but still accurate representation of the musculoskeletal system within shorter timeframes.
Hierarchically Structured Bioinspired Nanocomposites D. Nepal, S Kang, K. M. Adstedt, K. Kanhaiya, M. R. Bockstaller, L. C. Brinson, M. J. Buehler, P. V. Coveney, K. Dayal, J. A. El-Awady, L. C. Henderson, D. L. Kaplan, S. Keten, N. A. Kotov, G. C. Schatz, S. Vignolini, F. Vollrath, Y. Wang, B. I. Yakobson, V. V. Tsukruk, H. Heinz, (2022), DOI:10.1038/s41563-022-01384-1

Next-generation structural materials are expected to be lightweight, high-strength and tough composites with embedded functionalities to sense, adapt, self-repair, morph and restore. This Review highlights recent developments and concepts in bioinspired nanocomposites, emphasizing tailoring of the architecture, interphases and confinement to achieve dynamic and synergetic responses. We highlight cornerstone examples from natural materials with unique mechanical property combinations based on relatively simple building blocks produced in aqueous environments under ambient conditions. A particular focus is on structural hierarchies across multiple length scales to achieve multifunctionality and robustness. We further discuss recent advances, trends and emerging opportunities for combining biological and synthetic components, state-of-the-art characterization and modelling approaches to assess the physical principles underlying nature-inspired design and mechanical responses at multiple length scales. These multidisciplinary approaches promote the synergetic enhancement of individual materials properties and an improved predictive and prescriptive design of the next era of structural materials at multilength scales for a wide range of applications.

FabSim3: An automation toolkit for verified simulations using high performance computing D. Groen, H. Arabnejad, D. Suleimenova, W. Edeling, E. Raffin, Y. Xue, K. Bronik, N. Monnier, P. V. Coveney, Computer Physics Communications DOI:10.1016/j.cpc.2022.108596

Automation tools can help ensure the credibility of simulation results by reducing the manual time and effort required to perform these research tasks, by making more rigorous procedures tractable, and by reducing the probability of human error due to a reduced number of manual actions. In addition, efficiency gained through automation can help researchers to perform more research within the budget and effort constraints imposed by their projects. This paper presents the main software release of FabSim3, and explains how our automation toolkit can improve and simplify a range of tasks for researchers and application developers.

Parametric Analysis of an Efficient Boundary Condition to Control Outlet Flow Rates in Large Arterial Networks Sharp Chim Yui Lo, J. W. S. McCullough, P. V. Coveney, Sci. Rep., DOI:10.1038/s41598-022-21923-9 Substantial effort is being invested in the creation of a virtual human—a model which will improve our understanding of human physiology and diseases and assist clinicians in the design of personalised medical treatments. A central challenge of achieving blood flow simulations at full-human scale is the development of an efficient and accurate approach to imposing boundary conditions on many outlets. A previous study proposed an efficient method for implementing the two-element Windkessel model to control the flow rate ratios at outlets. Here we clarify the general role of the resistance and capacitance in this approach and conduct a parametric sweep to examine how to choose their values for complex geometries. We show that the error of the flow rate ratios decreases exponentially as the resistance increases. Our findings also establish constraints on the parameters controlling the numerical stability of the simulations. The findings from this work are directly applicable to larger and more complex vascular domains encountered at full-human scale.
Development and performance of a HemeLB GPU code for human-scale blood flow simulation I. Zacharoudiou, J.W.S. McCullough, P.V. Coveney, 282, 108548 (2023), Computer Physics Communications, DOI:10.21203/rs.3.rs-1305290/v1 In recent years, it has become increasingly common for high performance computers (HPC) to possess some level of heterogeneous architecture – typically in the form of GPU accelerators. In some machines these are isolated within a dedicated partition, whilst in others they are integral to all compute nodes – often with multiple GPUs per node – and provide the majority of a machine’s compute performance. In light of this trend, it is becoming essential that codes deployed on HPC are updated to execute on accelerator hardware. In this paper we introduce a GPU implementation of the 3D blood flow simulation code HemeLB that has been developed using CUDA C++. With HPC positioned on the brink of the exascale era, we use HemeLB as a motivation to provide a discussion on some of the challenges that many users will face when deploying their own applications on upcoming exascale machines
The performance of ensemble-based free energy protocols in computing binding affinities to ROS1 kinase S. Wan, A. P. Bhati, D. W. Wright, A. D. Wade, G. Tresaderm, H. van Vlijmen, P. V. Coveney, Sci. Rep., 12, 10433 (2022), doi: 10.1038/s41598-022-13319-6 Optimization of binding affinities for compounds to their target protein is a primary objective in drug discovery. Herein we report on a collaborative study that evaluates a set of compounds binding to ROS1 kinase. We use ESMACS (enhanced sampling of molecular dynamics with approximation of continuum solvent) and TIES (thermodynamic integration with enhanced sampling) protocols to rank the binding free energies. The predicted binding free energies from ESMACS simulations show good correlations with experimental data for subsets of the compounds. Consistent binding free energy differences are generated for TIES and ESMACS. Although an unexplained overestimation exists, we obtain excellent statistical rankings across the set of compounds from the TIES protocol, with a Pearson correlation coefficient of 0.90 between calculated and experimental activities.
Alchemical Free Energy Estimators and Molecular Dynamics Engines: Accuracy, Precision and Reproducibility A. Wade, A. P. Bhati, S. Wan, P. V. Coveney, J. Chem. Theory Comput., 18, 6, 3972–3987 (2022), doi: 10.1021/acs.jctc.2c00114 The binding free energy between a ligand and its target protein is an essential quantity to know at all stages of the drug discovery pipeline. Assessing this value computationally can offer insight into where efforts should be focused in the pursuit of effective therapeutics to treat a myriad of diseases. In this work, we examine the computation of alchemical relative binding free energies with an eye for assessing reproducibility across popular molecular dynamics packages and free energy estimators. The focus of this work is on 54 ligand transformations from a diverse set of protein targets: MCL1, PTP1B, TYK2, CDK2, and thrombin. These targets are studied with three popular molecular dynamics packages: OpenMM, NAMD2, and NAMD3 alpha. Agreement between thermodynamic integration and free energy perturbation is shown to be very good when using ensemble averaging.
Ensemble Simulations and Experimental Free Energy Distributions: Evaluation and Characterization of Isoxazole Amides as SMYD3 Inhibitors S. Wan, A. Bhati, D. Wright, I. Wall, A. Graves, D. Green, P. V. Coveney, J. Chem. Inf. Model., (2022), DOI: 10.1021/acs.jcim.2c00255 Optimization of binding affinities for ligands to their target protein is a primary objective in rational drug discovery. Herein, we report on a collaborative study that evaluates various compounds designed to bind to the SET and MYND domain-containing protein 3 (SMYD3). SMYD3 is a histone methyltransferase and plays an important role in transcriptional regulation in cell proliferation, cell cycle, and human carcinogenesis. Experimental measurements using the scintillation proximity assay show that the distributions of binding free energies from a large number of independent measurements exhibit non-normal properties. ESMACS and TIES are again found to be powerful protocols for the accurate comparison of the binding free energies.
Determining Clinically-Viable Biomarkers for Ischaemic Stroke Through a Mechanistic and Machine Learning Approach I. Benemerito, A. P. Narata, A. Narracott, A. Marzo, Annals of Biomedical Engineering, (2022), doi:10.1007/s10439-022-02956-7 Assessment of distal cerebral perfusion after ischaemic stroke is currently only possible through expensive and time-consuming imaging procedures which require the injection of a contrast medium. Alternative approaches that could indicate earlier the impact of blood flow occlusion on distal cerebral perfusion are currently lacking. The aim of this study was to identify novel biomarkers suitable for clinical implementation using less invasive diagnostic techniques such as Transcranial Doppler (TCD). We used 1D modelling to simulate pre- and post stroke velocity and flow wave propagation in a typical arterial network, and Sobol’s sensitivity analysis, supported by the use of Gaussian process emulators, to identify biomarkers linked to cerebral perfusion. We showed that values of pulsatility index of the right anterior cerebral artery > 1.6 are associated with poor perfusion and may require immediate intervention. Three additional biomarkers with similar behaviour, all related to pulsatility indices, were identified. These results suggest that flow pulsatility measured at specific locations could be used to effectively estimate distal cerebral perfusion rates, and ultimately improve clinical diagnosis and management of ischaemic stroke.
Personalized pathology test for Cardio-vascular disease: Approximate Bayesian computation with discriminative summary statistics learning R. Dutta., K. Z. Boudjeitia., C. Kotsalos., A. Rousseau., D. R. de Sousa., J. Desmet., A. Van Meerhaeghe., A. Mira., B. Chopard, PLOS Comput. Biol., (2022), doi:10.1371/journal.pcbi.1009910 Cardiovascular accidents often result from blood deficiencies, such as platelets dysfunction. Current diagnosis techniques to detect such dysfunctions are not sufficiently accurate and unable to determine which platelet properties are affected. We develop a novel approach to describe in-vitro platelets deposition patterns in terms of clinically meaningful patient specific bio-physical quantities that allow for personalized clinical diagnostics. This approach combines mathematical modeling, statistical inference techniques, machine learning and high performance computation to estimate the values of these clinically relevant platelet properties. We demonstrate our approach on three classes of donors, healthy volunteers, patients subject to dialysis and patients with chronic obstructive pulmonary disease. We claim that our approach opens a paradigm shift for the treatment and diagnosis of cardiovascular diseases, leading to personalized medicine.
Large Scale Study of Ligand-Protein Relative Binding Free Energy Calculations: Actionable Predictions from Statistically Robust Protocols A. P. Bhati and P. V. Coveney, J. Chem. Theory Comput., (2022), doi:10.1021/acs.jctc.1c01288 The accurate and reliable prediction of protein–ligand binding affinities can play a central role in the drug discovery process as well as in personalized medicine. Of considerable importance during lead optimization are the alchemical free energy methods that furnish an estimation of relative binding free energies (RBFE) of similar molecules. Recent advances in these methods have increased their speed, accuracy, and precision. This is evident from the increasing number of retrospective as well as prospective studies employing them. However, such methods still have limited applicability in real-world scenarios due to a number of important yet unresolved issues. Here, we report the findings from a large data set comprising over 500 ligand transformations spanning over 300 ligands binding to a diverse set of 14 different protein targets which furnish statistically robust results on the accuracy, precision, and reproducibility of RBFE calculations. Our findings provide a key set of recommendations that should be adopted for the reliable application of RBFE methods.
Hybrid parallelization of molecular dynamics simulations to reduce load imbalance J. Morillo, M. Vassaux, P. V. Coveney, M. Garcia-Gasulla, J. Supercomput. (2022), DOI:10.1007/s11227-021-04214-4 The most widely used technique to allow for parallel simulations in molecular dynamics is spatial domain decomposition, where the physical geometry is divided into boxes, one per processor. This technique can inherently produce computational load imbalance when either the spatial distribution of particles or the computational cost per particle is not uniform. This paper shows the benefits of using a hybrid MPI+OpenMP model to deal with this load imbalance. We consider LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator), a prototypical molecular dynamics simulator that provides its own balancing mechanism and an OpenMP implementation for many of its modules, allowing for a hybrid setup. In this work, we extend the current OpenMP implementation of LAMMPS and optimize it and evaluate three different setups: MPI-only, MPI with the LAMMPS balance mechanism, and hybrid setup using our improved OpenMP version.

2021

Title Citation Summary
An efficient, localised approach for the simulation of elastic blood vessels using the lattice Boltzmann method J. W. S. McCullough, P. V. Coveney, Scientific Reports (2021) DOI: 10.1038/s41598-021-03584-2 Introducing a novel approach to blood flow simulations, the study integrates elastic wall effects using the lattice Boltzmann method via the HemeLB open-source code. Unlike rigid-wall assumptions commonly employed, this method offers enhanced accuracy in capturing flow behavior in elastic vessels. Remarkably, it maintains computational efficiency and scalability on high-performance computers. Notably, the model accurately reproduces trends in wall shear stress distribution observed in studies coupling fluid dynamics and solid mechanics in personalized vascular geometries. These findings underscore the effectiveness of the proposed model in representing blood flow in elastic vessels, marking a potential leap forward in understanding cardiovascular dynamics.
Pandemic Drugs at Pandemic Speed: Accelerating COVID-19 Drug Discovery with Hybrid Machine Learning- and Physics-based Simulations on High Performance Computers A. P. Bhati, S. Wan, D. Alfè, A. R. Clyde, M. Bode, L. Tan, M. Titov, A. Merzky, M. Turilli, S. Jha, R. R. Highfield, W. Rocchia, N. Scafuri, S. Succi, D. Kranzlmüller, G. Mathias, D. Wifling, Y. Donon, A. Di Meglio, S. Vallecorsa, H. Ma, A. Trifan, A. Ramanathan, T. Brettin, A. Partin, F. Xia, X. Duan, R. Stevens, P. V. Coveney, Interface Focus (2021) DOI: 10.1098/rsfs.2021.0018 The global pandemic has underscored the inefficiencies of traditional drug discovery methods, prompting a search for faster and more cost-effective approaches. One significant challenge lies in screening numerous potential small molecules to identify lead compounds for antiviral drug development. This study proposes a novel solution by integrating machine learning methods, originally developed for linear accelerators, with physics-based methods. By combining these two in silico approaches, each with its own strengths and weaknesses, the research aims to accelerate drug discovery. The proposed workflow relies on supercomputing to achieve exceptionally high throughput. The study demonstrates the feasibility of this approach by applying it to the study of inhibitors for four COVID-19 target proteins. Utilizing various supercomputers, the researchers successfully conduct large-scale calculations to identify lead antiviral compounds through repurposing. This innovative infrastructural development offers promising prospects for expediting drug discovery processes and addressing urgent global health challenges.
Possible Contexts of Use for In Silico Trials Methodologies: A Consensus-Based Review Marco Viceconti; Luca Emili; Payman Afshari; Eulalie Courcelles; Cristina Curreli; Nele Famaey; Liesbet Geris; Marc Horner; Maria Cristina Jori; Alexander Kulesza; Axel Loewe; Michael Neidlin; Markus Reiterer; Cécile F. Rousseau; Giulia Russo; Simon J. Sonntag; Emmanuelle M. Voisin; Francesco Pappalardo, IEEE Journal of Biomedical and Health Informatics (2021) DOI: 10.5445/ir/1000137059 “In Silico Trials” refer to the utilization of computer modeling and simulation to assess the safety and effectiveness of various medical products, including drugs, medical devices, diagnostics, and advanced therapy medicinal products. These predictive models are emerging as novel methodologies for both the development and regulatory evaluation of medical products. Regulatory bodies such as the FDA and EMA qualify new methodologies through formal processes, starting with the definition of the Context of Use (CoU), which outlines how the new methodology will be applied in the development and regulatory assessment process. Given the disruptive nature of In Silico Trials, establishing a list of potential CoUs is crucial to delineate their applications for regulatory science development. This review paper presents the outcomes of a consensus process conducted within the InSilicoWorld Community of Practice, an online platform for experts in in silico medicine. The experts identified 46 descriptions of potential CoUs, which were categorized into nine CoU categories. While examples of 31 CoUs were found in the literature, the remaining 15 are currently speculative and require further exploration.
Time-efficient three-dimensional transmural scar assessment provides relevant substrate characterization for ventricular tachycardia features and long-term recurrences in ischemic cardiomyopathy. Susana Merino-Caviedes; Lilian K. Gutiérrez; José Manuel Alfonso-Almazán; Santiago Sanz-Estébanez; Lucilio Cordero-Grande; Jorge G. Quintanilla; Jorge G. Quintanilla; Javier Sánchez-González; Manuel Marina-Breysse; Carlos Galán-Arriola; Daniel Enríquez-Vázquez; Daniel Enríquez-Vázquez; Carlos Torres; Gonzalo Pizarro; Borja Ibanez; Rafael Peinado; José L. Merino; Julián Pérez-Villacastín; José Jalife; Mariña López-Yunta; Mariano Vázquez; Jazmin Aguado-Sierra; Juan José González-Ferrer; Nicasio Pérez-Castellano; Marcos Martín-Fernández; Carlos Alberola-López; David Filgueiras-Rama; David Filgueiras-Rama, Scientific Reports (2021) DOI: 10.1038/s41598-021-97399-w In patients with ischemic cardiomyopathy and ventricular arrhythmia, delayed gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) imaging presents challenges in efficiently characterizing myocardial substrate. This study introduces a novel 3D methodology, validated in both pigs and patients with myocardial infarction, to assess ventricular scar. Custom transmural criteria and a semiautomatic approach are utilized to generate transmural scar maps in ventricular models reconstructed from both 3D-acquired and 3D-upsampled-2D-acquired LGE-CMR images. Results indicate that 3D-upsampled models from 2D LGE-CMR images offer a time-efficient alternative to 3D-acquired sequences for assessing myocardial substrate. Scar assessment using 3D-upsampled models from 2D-LGE-CMR sequences outperforms conventional 2D assessment in identifying scar sizes associated with spontaneous ventricular tachycardia episodes’ cycle length and long-term ventricular tachycardia recurrences after catheter ablation. This novel methodology, following manual or automatic segmentation of myocardial borders in a small number of conventional 2D LGE-CMR slices and automatic scar detection, could prove to be an efficient approach in clinical practice.
Thermodynamic and structural insights into the repurposing of drugs that bind to SARS-CoV-2 main protease S. Wan, A. Bhati, A. Wade, D. Alfe, P. V. Coveney, Molecular Systems Design & Engineering (2021) DOI: 10.1039/d1me00124h Researchers have been diligently working since the COVID-19 outbreak, resulting in the approval of only three drugs – remdesivir, ronapreve, and molnupiravir – which directly target the SARS-CoV-2 virus in certain countries. Due to the slow pace and high costs of new drug discovery and development, coupled with the urgency of the situation, repurposing existing drugs for COVID-19 treatment is an appealing option. A recent study conducted a high-throughput X-ray crystallographic screen for a selection of drugs approved or in clinical trials. Thirty-seven compounds were identified from drug libraries, all binding to the SARS-CoV-2 main protease (3CLpro). In this study, molecular dynamics simulation and an ensemble-based free energy approach, ESMACS (enhanced sampling of molecular dynamics with approximation of continuum solvent), were employed to investigate a subset of these compounds. The drugs studied exhibit high diversity, interacting with various binding sites and/or subsites of 3CLpro. Predicted free energies are compared with experimental results, showing excellent agreement. Additionally, the study provides detailed energetic insights into the drug-protein binding, facilitating the design and discovery of potential drugs for COVID-19 treatment.
A completely automated pipeline for 3D reconstruction of human heart from 2D cine magnetic resonance slices. Banerjee, A; Camps, J; Zacur, E; Andrews, CM; Rudy, Y; Choudhury, RP; Rodriguez, B; Grau, V, Philosophical Transactions of the Royal Society A (2021) DOI: 10.1098/rsta.2020.0257 Cardiac magnetic resonance (CMR) imaging is vital for diagnosing cardiovascular diseases non-invasively, yet its clinical acquisition as separate 2D image planes limits 3D analysis accuracy. This paper introduces an automated pipeline for generating patient-specific 3D biventricular heart models from cine MR slices. The pipeline selects relevant cine MR images, segments them using deep learning-based methods, and aligns contours in 3D space to correct misalignments. A statistical shape model aids in alignment. Finally, a sparse 3D representation of contours generates a smooth 3D biventricular mesh. Applied to a CMR dataset of 20 healthy subjects, the pipeline reduces misalignment artifacts on average from 1.82 ± 1.60 mm to 0.72 ± 0.73 mm across subjects. High-resolution 3D biventricular meshes enable simulations of electrical activation patterns, showing agreement with non-invasive electrocardiographic imaging. These automated methodologies advance precision medicine, enhancing clinical data interpretability, patient-specific modeling, and simulation for digital twin vision and augmented reality applications.
Cardiac computational modelling Bragard, Jean R.; Cámara, Óscar; Echebarria, Blas; Giorda, Luca Gerardo; Pueyo, Esther; Saiz Rodríguez, Francisco Javier; Sebastián, Rafael; Soudah, Eduardo; Vázquez, Mariano, Revista Española de Cardiología (English Edition) (2021) DOI: 10.1016/j.rec.2020.05.024 Cardiovascular diseases pose significant social and economic challenges, ranking among the leading causes of mortality and morbidity worldwide. Personalized computational models of the heart are proving invaluable in elucidating disease mechanisms, optimizing treatment strategies, and predicting patient responses. To advance this field, the Spanish Research Network for Cardiac Computational Modelling (VHeart-SN) has been established. The network aims to develop an integrated, modular, and multiscale multiphysical computational model of the heart. Specific objectives include integrating diverse numerical methods and models tailored to individual patients, advancing understanding of cardiac and vascular disease mechanisms, and facilitating the application of personalized therapies. This article outlines the current state of cardiac computational modelling and highlights various scientific endeavors undertaken by network members to explore the features and utility of these models.
Concomitant Respiratory Failure Can Impair Myocardial Oxygenation in Patients with Acute Cardiogenic Shock Supported by VA-ECMO Anthony R. Prisco; Jazmin Aguado-Sierra; Constantine Butakoff; Mariano Vázquez; Guillaume Houzeaux; Beatriz Eguzkitza; Jason A. Bartos; Demetris Yannopoulos; Ganesh Raveendran; Mikayle A. Holm; Tinen L. Iles; Claudius Mahr; Paul A. Iaizzo, Journal of Cardiovascular Translational Research (2021) DOI: 10.1007/s12265-021-10110-2 Venous-arterial extracorporeal membrane oxygenation (VA-ECMO) therapy in acute cardiogenic shock with concomitant acute lung injury can lead to a severe complication known as “north-south syndrome” (NSS), resulting in cerebral hypoxia. However, NSS remains poorly understood, with hemodynamic studies focusing mainly on cerebral perfusion and neglecting cardiac aspects. In this study, we aimed to investigate whether the heart is more susceptible to hypoxemic blood flow than the brain during NSS, considering the proximity of coronary arteries to the aortic annulus. Computational fluid dynamics simulations of blood flow in a human supported by VA-ECMO were conducted. The simulations quantified the proportion of blood at each aortic branching vessel originating from residual native cardiac output versus VA-ECMO. Results revealed that as residual cardiac function increased, myocardial hypoxia would occur before cerebral hypoxia, shedding light on the conditions leading to NSS development and the relative cardiac function associated with organ-specific hypoxia.
Design and execution of a Verification, Validation, and Uncertainty Quantification plan for a numerical model of left ventricular flow after LVAD implantation Alfonso Santiago; Constantine Butakoff; Beatriz Eguzkitza; Richard A. Gray; Karen May-Newman; Pras Pathmanathan; Vi Vu; Mariano Vázquez, PLOS Computational Biology (2021) DOI: 10.1371/journal.pcbi.1010141 During medical device regulatory evaluations, manufacturers traditionally provide evidence of safety and efficacy through bench, animal, and human trials. Yet, as inquiries become more complex, traditional methods struggle to deliver reliable answers. Numerical modeling offers a solution but faces challenges in practical application due to deterministic nature and uncertainties. ASME V&V40 standardizes credibility assessment for simulation results, emphasizing thorough model checks and comparison with physical experiments. This manuscript demonstrates a VVUQ process following ASME V&V40, ensuring robustness in evaluating medical device simulations.
Effect of Suboptimal Neuromuscular Control on the Risk of Massive Wear in Total Knee Replacement Marco Viceconti; Cristina Curreli; Francesca Bottin; Giorgio Davico, Annals of Biomedical Engineering (2021) DOI: 10.1007/s10439-021-02795-y During level walking, optimal neuromuscular control minimizes metabolic energy consumption and joint force transmission. Patients with total knee replacements, especially those with Parkinson’s Disease (PD), often adopt suboptimal control strategies, increasing joint loading and implant failure risk. We used a biomechanical model to estimate the impact of severely suboptimal control on total knee replacement wear rates. While non-PD patients typically experience 17 mm3 annual polyethylene wear, simulations suggested up to 69 mm3 with suboptimal control, potentially increasing implant failure risk from 0.1% to 0.5%. However, this alone may not explain higher revision rates in PD patients, implying other factors contribute to failure.
Implementation Details of the Reconstruction Pipeline and Electrophysiological Inference Results from A completely automated pipeline for 3D reconstruction of human heart from 2D cine magnetic resonance slices Banerjee, Abhirup; Camps, Julià; Zacur, Ernesto; Andrews, Christopher M.; Rudy, Yoram; Choudhury, Robin P.; Rodriguez, Blanca; Grau, Vicente, Figshare (2021) DOI: 10.6084/m9.figshare.15656924.v2 The paper introduces an automated pipeline for creating patient-specific 3D biventricular heart models from cine magnetic resonance (MR) slices. Traditionally, CMR imaging, a valuable tool in cardiovascular disease diagnosis, acquires 2D image planes separately, limiting 3D analysis accuracy. The proposed pipeline addresses this limitation by automatically selecting relevant cine MR images, segmenting them with a deep learning-based method to extract heart contours, and aligning the contours in 3D space. This alignment corrects potential misalignments due to breathing or subject motion using both cine data intensity and contours information, as well as a statistical shape model.
Modelización computacional cardiaca Jean Bragard; Oscar Camara; Blas Echebarria; Luca Gerardo Giorda; Esther Pueyo; Javier Saiz; Rafael Sebastian; Eduardo Soudah; Mariano Vázquez, Revista Española de Cardiología (2021) DOI: 10.1016/j.recesp.2020.05.040 Cardiovascular diseases exert significant social and economic burdens, ranking among the leading causes of mortality and morbidity worldwide. Personalized computational models of the heart are emerging as valuable tools for elucidating the mechanisms underlying cardiac pathologies and optimizing treatment strategies by predicting individual patient responses. In this context, the Spanish Network for Research in Cardiac Computational Modeling (V-Heart SN) has been established. Its overarching aim is to develop an integrated, multiphysical, and multiscale computational model of the heart. Specific objectives include integrating diverse numerical models to account for patient-specificity, advancing understanding of cardiac and vascular pathologies, and supporting the application of personalized therapies. This article outlines the current status of cardiac computational modeling and highlights various scientific contributions from network members aimed at enhancing understanding of model characteristics and utility.
Palabos-npFEM: Software for the Simulation of Cellular Blood Flow (Digital Blood) Christos Kotsalos; Jonas Latt; Bastien Chopard, Journal of Open Research Software (2021) DOI: 10.5334/jors.343 Palabos-npFEM is a computational framework designed for simulating blood flow with fully resolved constituents. This software enables the resolution of trajectories and the deformed state of blood cells, including red blood cells and platelets, along with their complex interactions. The tool combines multiple solvers: the lattice Boltzmann solver Palabos for simulating blood plasma (fluid phase), a finite element method (FEM) solver for resolving blood cells (solid phase), and an immersed boundary method (IBM) for coupling the two phases. Palabos-npFEM offers both CPU-only and GPU-accelerated versions, making it suitable for deployment on high-performance computing systems. The software is integrated into the Palabos core library and is accessible via the Git repository at https://gitlab.com/unigespc/palabos. It supports various simulation setups, including different geometries and blood parameters. Additionally, due to its modular design, external solvers can be easily incorporated to replace the provided ones.
Predicting Residence Time of GPCR Ligands with Machine Learning. Potterton A, Heifetz A, Townsend-Nicholson A, Methods in Molecular Biology (2021) DOI: 10.1007/978-1-0716-1787-8_8 Drug-target residence time, the duration of binding to a protein target, is increasingly recognized for its importance in drug efficacy. However, predicting residence time presents challenges due to limited data availability. This chapter addresses this issue by compiling the largest publicly available repository of GPCR-ligand kinetic data. It summarizes experimental evidence on factors influencing residence time and outlines two machine learning workflows for prediction: a single-target model based on ligand features, and a multi-target model utilizing features from molecular dynamics simulations. These efforts aim to facilitate efficient optimization of residence time in drug discovery by deciphering influential kinetic data features for computational modeling.
The Effects of Micro-vessel Curvature Induced Elongational Flows on Platelet Adhesion Spieker, C.J., Závodszky, G., Mouriaux, C., van der Kolk, M., Gachet, C., Mangin, P.H., Hoekstra, A.G., Annals of Biomedical Engineering (2021) DOI: 10.1007/s10439-021-02870-4 The study explores the influence of channel curvature on cellular blood flow, particularly platelet adhesion, a crucial aspect in various diseases and physiological processes. Simulations in a half-arc channel reveal significant differences in shear rate distribution between inner and outer arcs, while cell distributions remain largely unaffected. Experimental validation shows elevated platelet adhesion on the inner arc at high shear rates, correlating with simulation predictions. Platelet availability for binding appears unaffected by curvature, suggesting mechanical rather than probabilistic influence on binding. The presence of elongational flows in simulations highlights their potential role in increased platelet adhesion, contributing to our understanding of blood flow dynamics in curved geometries.
Three-dimensional analysis of blood platelet spreading using digital holographic microscopy: a statistical study of the differential effect of coatings in healthy volunteers and dialyzed patients Jérôme Dohet-Eraly; Karim Zouaoui Boudjeltia; Alexandre Rousseau; Patrick Queeckers; Christophe Lelubre; Jean-Marc Desmet; Bastien Chopard; Catherine Yourassowsky; Frank Dubois, Biomedical Optics Express (2021) DOI: 10.1364/boe.448817 This paper utilizes interferometric digital holographic microscopy to measure single deposited platelets, crucial in cardiovascular disorders due to their role in blood clotting. Comparing platelets from healthy volunteers to dialyzed patients, the study reveals significantly lower average optical height in healthy individuals, indicating better spreading. This underscores the importance of assessing platelet parameters in patients and highlights the potential of digital holographic microscopy for fundamental research and diagnostic applications in routine laboratories, enhancing traditional blood tests.
A systematic approach to the scale separation problem in the development of multiscale models Bhattacharya, P., Li, Q., Lacroix, D., Kadirkamanathan, V., Viceconti, M., PLOS ONE (2021) DOI: 10.1371/journal.pone.0251297 Addressing multiscale challenges in engineering, particularly in biomedical contexts, this work proposes a novel approach to defining scale. By considering measurement limitations, computational capacity, and characteristic variations of quantities over time and space, the study develops a comprehensive multiscale modeling methodology. This methodology, illustrated with a specific problem, offers adaptability to similar engineering challenges. While the paper focuses on a particular example, its proposed approach holds potential for broader application across diverse engineering domains.
Applying the CiPA approach to evaluate cardiac proarrhythmia risk of some antimalarials used off-label in the first wave of COVID-19 Annie Delaunois, Matthew Abernathy, Warren D. Anderson, Kylie A. Beattie, Khuram W. Chaudhary, Julie Coulot, Vitalina Gryshkova, Simon Hebeisen, Mark Holbrook, James Kramer, Yuri Kuryshev, Derek Leishman, Isabel Lushbough, Elisa Passini, Will S. Redfern, Blanca Rodriguez, Eric I. Rossman, Cristian Trovato, Caiyun Wu, Jean-Pierre Valentin, Clinical and Translational Science (2021) DOI: 10.1111/cts.13011 We conducted in silico and in vitro assays following the Comprehensive In Vitro Proarrhythmia Assay (CiPA) paradigm to evaluate the risk of chloroquine (CLQ) or hydroxychloroquine (OH-CLQ)-mediated QT prolongation and Torsades de Pointes (TdP). These drugs, repurposed during the COVID-19 pandemic, were tested alone and combined with erythromycin (ERT) and azithromycin (AZI). Patch clamp assays on cardiac ion channels, in silico models of human ventricular electrophysiology, and human-induced pluripotent stem cell (hiPSC)-derived cardiomyocytes were employed. CLQ and OH-CLQ showed inhibitory effects on potassium, sodium, and calcium currents, particularly on IKr and IK1. Virtual Assay indicated TdP risk, more pronounced in high-risk populations. All drugs induced early after-depolarizations in hiPSC-derived cardiomyocytes, with no exacerbation seen with macrolide combinations.
Biventricular Surface Reconstruction From Cine Mri Contours Using Point Completion Networks Beetz, Marcel, Abhirup Banerjee, and Vicente Grau, IEEE 18th International Symposium on Biomedical Imaging (ISBI) (2021) DOI: 10.1109/ISBI48211.2021.9434040 This paper introduces a novel deep learning approach for reconstructing dense 3D biventricular heart models from misaligned 2D cardiac MR image contours. Traditional MRI protocols often produce 2D slices of 3D anatomy, prone to motion artifacts and misalignment. The proposed method directly operates on point clouds, effectively reducing slice misalignments and producing smooth 3D shapes with accurate left ventricular volumes. Tested on datasets with varying degrees of misalignment, it demonstrates robustness and achieves high accuracy, showcasing its potential for improving cardiac MRI analysis and clinical practice.
Comparison of the Simulated Response of Three in Silico Human Stem Cell-Derived Cardiomyocytes Models and in Vitro Data Under 15 Drug Actions Michelangelo Paci, Jussi T. Koivumäki, Hua Rong Lu, David J. Gallacher, Elisa Passini, and Blanca Rodriguez, Frontiers in Pharmacology (2021) DOI: 10.3389/fphar.2021.604713 Improvements in hSC-CM technology led to their increased use in drug testing. This study assessed the response of three in silico hSC-CM models to simulated drug action and compared results against in vitro data for 15 drugs. Simulations analyzed CTD90 changes and arrhythmic events, showing overall consistency among models but differences compared to in vitro measurements. Arrhythmic events varied across models, influenced by drug effects on ion channels. Spontaneous activity suppression also varied. While in silico CTD90 changes were consistent with in vitro data, differences in drug responses across models may stem from variability in experimental data used in their construction.
des-ist: A Simulation Framework to Streamline Event-Based In Silico Trials M. van der Kolk, C. Miller, R. Padmos, V. Azizi, A. Hoekstra, Computational Science (2021) DOI: 10.1007/978-3-030-77967-2_53 To promote the use of in silico trials for medical device, drug, or treatment development, we introduce the des-ist framework (Discrete Event Simulation framework for In Silico Trials). It facilitates discrete event-based simulations organized in an acyclic, directed graph, where each node represents a component of the trial. With a simple API and data layout, des-ist allows easy coupling of numerous simulations via containerized environments like Docker and Singularity. An acute ischemic stroke treatment trial, as in the INSIST project, illustrates its application. des-ist streamlines model coupling, ensures reproducibility through containerization, supports parallel execution via GNU Parallel, and aids in managing large virtual cohorts. Future enhancements aim to integrate validation, verification, and uncertainty quantification analyses for improved sensitivity analysis and trust in computational outcomes, expediting treatment development.
Enhanced single-node lattice Boltzmann boundary condition for fluid flows Francesco Marson, Yann Thorimbert, Bastien Chopard, Irina Ginzburg, and Jonas Latt, Physical Review E (2021) DOI: 10.1103/PhysRevE.103.053308 We introduce a novel approach to implement Dirichlet velocity boundary conditions for complex shapes within the lattice Boltzmann method, utilizing data from a single node. This method combines the bounce-back rule with interpolations, enhancing accuracy by restricting interpolation extension near the boundary. Despite its local nature, the approach demonstrates second-order convergence for velocity fields and comparable or superior accuracy to established schemes for curved walls. We identify meaningful variants of the method distinguished by their approximation of second-order nonequilibrium terms and interpolation coefficients, each offering viscosity-independent accuracy. This method is versatile, applicable to moving rigid surfaces with various kinematics and suitable for simulation across diverse geometries, including singular ones.
FAIR Principles for Research Software (FAIR4RS Principles) Chue Hong, N. P., Katz, D. S., Barker, M., Lamprecht, A.-L., Martinez, C., Psomopoulos, F. E., Harrow, J., Castro, L. J., Gruenpeter, M., Martinez, P. A., Honeyman, T., et al., Research Data Alliance (2021) DOI: 10.15497/RDA00065 Research software plays a critical role in scholarly endeavors, yet it faces challenges related to discoverability, productivity, quality, reproducibility, and sustainability. To address these issues, the open science, open source, and FAIR (Findable, Accessible, Interoperable, and Reusable) communities have converged to advocate for the application of FAIR principles to research software. This recognition marks a significant step forward in acknowledging the importance of FAIR research software in enhancing the value of research outcomes. The FAIR for Research Software (FAIR4RS) Working Group has developed the FAIR Principles for Research Software (FAIR4RS Principles), which adapt the FAIR Guiding Principles to the realm of research software. This adaptation aims to facilitate discussions on the adoption of FAIR principles and promote their implementation. Additionally, examples of organizations implementing these principles are provided to facilitate knowledge sharing and encourage broader adoption, thereby increasing the impact of research software.
High fidelity blood flow in a patient-specific arteriovenous fistula J. W. S. McCullough and P. Coveney, Scientific Reports (2021) DOI: 10.1038/s41598-021-01435-8 Utilizing supercomputing infrastructure, simulations are increasingly feasible for optimizing arteriovenous fistulae (AVF) placement in kidney dialysis patients. A 3D AVF model employing lattice Boltzmann method accurately resolves blood flow in patient-specific vasculature, covering the entire forearm. These simulations, validated against clinical data, mark a significant step towards personalized AVF planning. Future endeavors aim to incorporate intricate biophysics in real-time physiological models, offering broader applications in both basic research and clinical practice.
Human biventricular electromechanical simulations on the progression of electrocardiographic and mechanical abnormalities in post-myocardial infarction Wang, Z. J., Santiago, A., Zhou, X., Wang, L., Margara, F., Levrero-Florencio, F., Das, A., Kelly, C., Dall’Armellina, E., Vazquez, M., & Rodriguez, B., EP Europace (2021) DOI: 10.1093/europace/euaa405 Aiming to develop, calibrate, and evaluate a human electromechanical modelling and simulation framework for healthy and post-myocardial infarction (MI) conditions, this study conducted simulations with a novel biventricular model. It integrated experimental and clinical data, including torso/biventricular anatomy from clinical magnetic resonance, human-based membrane kinetics, and electromechanical coupling. Results were compared with clinical data, showing healthy model simulations with LVEF of 63% and specific electrocardiogram parameters. Post-MI states exhibited varying LVEF values, with mechanistic insights revealing pathophysiological changes in infarcted regions. The framework offers a testbed for optimizing therapies.
In Silico Trials for Treatment of Acute Ischemic Stroke: Design and Implementation Claire Miller, Raymond M. Padmos, Max van der Kolk, Tamás I. Józsa, Noor Samuels, Yidan Xue, Stephen J. Payne, Alfons G. Hoekstra, Computers in Biology and Medicine (2021) DOI: 10.1016/j.compbiomed.2021.104802 This paper introduces an in silico trial framework for acute ischemic stroke treatment, aiming to refine, reduce costs, and partially replace in vivo studies. Employing an event-based modelling approach, instantaneous changes in system states simulate disease progression and treatment effects. The modular design facilitates development and reproducibility across diverse models, covering multiple time scales. Statistical population and outcome models generate cohorts of virtual patients, predicting functional outcomes based on treatment and injury results. Proof of concept trials demonstrate the framework’s functionality, simulating successful and unsuccessful treatment scenarios. Challenges, including validation and computational constraints, are addressed to enhance trial setup.
Inference of ventricular activation properties from non-invasive electrocardiography Camps, Julia, Brodie Lawson, Christopher Drovandi, Ana Minchole, Zhinuo Jenny Wang, Vicente Grau, Kevin Burrage, and Blanca Rodriguez, Medical Image Analysis (2021) DOI: 10.1016/j.media.2021.102143 Precision cardiology necessitates innovative approaches for non-invasive patient-specific cardiac function assessment. Electrocardiography and imaging aid clinical diagnosis, yet integrating multi-modal data through advanced computational methods could yield a cardiac ‘digital twin’, revealing a patient’s heart condition and simulating treatment outcomes. This study introduces computational techniques to estimate key ventricular activation properties for individual subjects by combining electrocardiography, cardiac magnetic resonance imaging, and modeling. The approach utilizes a sequential Monte Carlo method with Eikonal simulations and torso-biventricular models derived from clinical imaging. Results demonstrate successful inference of ventricular activation properties, crucial for personalized cardiac assessments.
Left Ventricle Quantification Challenge: A Comprehensive Comparison and Evaluation of Segmentation and Regression for Mid-ventricular Short-axis Cardiac MR Data Xue, Wufeng, Jiahui Li, Zhiqiang Hu, Eric Kerfoot, James Clough, Ilkay Oksuz, Hao Xu, Vicente Grau, Fumin Guo, and Matthew Ng., IEEE Journal of Biomedical and Health Informatics (2021) DOI: 10.1109/JBHI.2021.3064353 Automatic quantification of the left ventricle (LV) from cardiac magnetic resonance (CMR) images is crucial for efficient diagnosis, yet lacks a systematic benchmarking platform due to variations in label information. This paper evaluates LV quantification methods from the LVQuan challenge at MICCAI 2018. Utilizing the Cardiac-DIG dataset with ground truth labels, submissions were assessed for areas, dimensions, regional wall thicknesses (RWT), and cardiac phase estimation. Both segmentation-based (SG) and direct regression (DR) methods showed promising performance, with the DR method LDAMT achieving the best results. The study discusses the strengths, weaknesses, and future directions of SG and DR methods for clinical application.
Optimised Misalignment Correction from Cine MR Slices Using Statistical Shape Model Banerjee, Abhirup, Ernesto Zacur, Robin P Choudhury, and Vicente Grau, Annual Conference on Medical Image Understanding and Analysis (2021) DOI: 10.1007/978-3-030-80432-9_16 Cardiac magnetic resonance (CMR) imaging offers valuable insights into cardiovascular diseases, yet its 3D reconstruction from 2D slices is hindered by misalignments. This study presents a method using a statistical shape model (SSM) to correct slice misalignments for accurate 3D heart modeling. Initial corrections based on image intensities and heart contours are followed by SSM-based alignment accounting for both in-plane and out-of-plane misalignments. Evaluation on a cohort from the UK Biobank shows a significant reduction in misalignment artifacts, improving from mm to mm distance in the final 3D mesh reconstruction.
Palabos: Parallel Lattice Boltzmann Solver Jonas Latt, Orestis Malaspinas, Dimitrios Kontaxakis, Andrea Parmigiani, Daniel Lagrava, Federico Brogi, Mohamed Ben Belgacem, Yann Thorimbert, Sébastien Leclaire, Sha Li, Francesco Marson, Jonathan Lemus, Christos Kotsalos, Raphaël Conradin, Christophe Coreixas, Rémy Petkantchin, Franck Raynaud, Joël Beny, Bastien Chopard, Computers & Mathematics with Applications (2021) DOI: 10.1016/j.camwa.2020.03.022 This article introduces Palabos, an open-source Lattice Boltzmann library designed for Computational Fluid Dynamics simulations since 2010. Palabos, developed in C++, aims at addressing complex, coupled physics in various applications. It offers a versatile modeling framework while maintaining strong computational performance. The paper outlines Palabos’ scope, concepts, data structures, and programming models, alongside detailing its philosophy and listing implemented models. Benchmark simulations validate core functionalities, affirming Palabos’ quality and suitability for diverse applications within the Lattice Boltzmann community.
Stochastic PCA-Based Bone Models from Inverse Transform Sampling: Proof of Concept for Mandibles and Proximal Femurs Pascoletti, G., Aldieri, A., Terzini, M., Bhattacharya, P., Calì, M., Zanetti, E.M., Applied Sciences (2021) DOI: 10.3390/app11115204 Principal components analysis (PCA) is applied to three-dimensional bone shape models, allowing the generation of numerous models from a limited set of variables. This study details the procedure and tests it using databases of mandibles (40 samples) and proximal femurs (98 samples). For both bones, the “average shape” and principal components covering 90% of variance are identified, along with the statistical distributions of component weights. Fifteen components describe mandibular shape, while nine suffice for proximal femur morphology. A routine is established to create geometries based on actual statistical shape distributions, applicable to other bone shapes with adequate data.
Tutorial applications for Verification, Validation and Uncertainty Quantification using VECMA toolkit D. Suleimenova, H. Arabnejad, W. N. Edeling, D. Coster, O. O. Luk, J. Lakhlili, V. Jancaskas, M Kulczewski, L. Veen, D. Ye, P. Zun, V. Krzhizhanovskaya, A. Hoekstra, D. Crommelin, P. V. Coveney, D. Groen, Journal of Computational Science (2021) DOI: 10.1016/j.jocs.2021.101402 The VECMA toolkit offers automated Verification, Validation, and Uncertainty Quantification (VVUQ) for diverse applications, deployable on emerging exascale platforms across domains. Comprising EasyVVUQ, FabSim3, MUSCLE3, and QCG tools, it facilitates VVUQ workflows, automation, multiscale model coupling, and HPC execution. Additionally, EasySurrogate supports various surrogate methods. This paper presents five tutorials across different domains utilizing these components for uncertainty quantification, surrogate modeling, multiscale coupling, and sensitivity analysis on HPC. Aimed at practitioners seeking hands-on experience, these tutorials provide a practical insight into utilizing VECMA toolkit functionalities for their applications.
Uncertainty Quantification of Coupled 1D Arterial Blood Flow and 3D Tissue Perfusion Models Using the INSIST Framework C. Miller, M. van der Kolk, R. Padmos, T. Józsa, A. Hoekstra, Computational Science (2021) DOI: 10.1007/978-3-030-77980-1_52 Uncertainty quantification is conducted on a one-dimensional arterial blood flow model, examining its impact on a coupled tissue perfusion model of the brain, focusing on acute ischemic stroke. Infarct volume, a key outcome, is estimated based on perfusion changes between healthy and occluded states. Secondary outcomes include uncertainty in blood flow at network outlets, affecting boundary conditions in the brain tissue perfusion model. Factors such as heart stroke volume, heart rate, blood density, and viscosity are considered. While outlet blood flow uncertainty aligns with input uncertainty, infarct volume uncertainty is notably smaller. These findings bolster the credibility of coupled models for in silico clinical trials.
High fidelity blood flow in a patient‑specific arteriovenous fistula J. W. S. McCullough and P. Coveney, Sci. Rep. 11:22301 (2021) DOI: 10.1038/s41598-021-01435-8 An arteriovenous fistula, created by artificially connecting segments of a patient’s vasculature, is the preferred way to gain access to the bloodstream for kidney dialysis. The increasing power and availability of supercomputing infrastructure means that it is becoming more realistic to use simulations to help identify the best type and location of a fistula for a specific patient. We describe a 3D fistula model that uses the lattice Boltzmann method to simultaneously resolve blood flow in patient‑specific arteries and veins. The simulations conducted here, comprising vasculatures of the whole forearm, demonstrate qualified validation against clinical data. Ongoing research to further encompass complex biophysics on realistic time scales will permit the use of human‑scale physiological models for basic and clinical medicine.
Principles of Small-Molecule Transport through Synthetic Nanopores T. Diederichs, K. Ahmed, J. Burns, Q. Nguyen, Z. Siwy, M. Tornow, P. Coveney, R. Tampé, S. Howorka, ACS Nano (2021), DOI: DOI:10.1021/acsnano.1c05139 Synthetic nanopores made from DNA replicate the key biological processes of transporting molecular cargo across lipid bilayers. Understanding transport across the confined lumen of the nanopores is of fundamental interest and of relevance to their rational design for biotechnological applications. Here we reveal the transport principles of organic molecules through DNA nanopores by synergistically combining experiments and computer simulations. Using a highly parallel nanostructured platform, we synchronously measure the kinetic flux across hundreds of individual pores to obtain rate constants. Our findings on these synthetic pores’ structure–function relationship will serve to guide their rational engineering to tailor transport selectivity for cell biological research, sensing, and drug delivery.
Delivering computationally-intensive digital patient applications to the clinic: An exemplar solution to predict femoral bone strength from CT data I. Benemerito, W. Griffiths, J. Allsopp, W. Furnass, P≥ Bhattacharya, X. Li, A. Marzo, S. Wood, M. Viceconti, A. Narracott, CMPB (2021), DOI: DOI:10.1016/j.cmpb.2021.106200 Whilst fragility hip fractures commonly affect elderly people, often causing permanent disability or death, they are rarely addressed in advance through preventive techniques. Quantification of bone strength can help to identify subjects at risk, thus reducing the incidence of fractures in the population. In recent years, researchers have shown that finite element models (FEMs) of the hip joint, derived from computed tomography (CT) images, can predict bone strength more accurately than other techniques currently used in the clinic. The specialised hardware and trained personnel required to perform such analyses, however, limits the widespread adoption of FEMs in clinical contexts. In this manuscript we present CT2S (Computed Tomography To Strength), a system developed in collaboration between The University of Sheffield and Sheffield Teaching Hospitals, designed to streamline access to this complex workflow for clinical end-users. We conclude that the short processing time makes the system compatible with current clinical workflows. The use of open source software and the accurate description of the workflow given here facilitates the deployment in other centres.
IMPECCABLE: Integrated Modeling PipelinE for COVID Cure by Assessing Better LEads A. Al Saadi, D. Alfe, Y. Babuji, A. Bhati, B. Blaiszik, A. Brace, T. Brettin, K. Chard, R. Chard, A. Clyde, P. V. Coveney, I. Foster, T. Gibbs, S. Jha, K. Keipert, T. Kurth, D. Kranzlmüller, H. Lee, Z. Li, H. Ma, A. Merzky, G. Mathias, A. Partin, J. Yin, A. Ramanathan, A. Shah, A. Stern, R. Stevens, L. Tan, M. Titov, A. Trifan, A. Tsaris, M. Turilli, H. Van Dam, S. Wan, D. Wifling, 50th International Conference on Parallel Processing (ICPP ’21), August 9-12 (2021), DOI: DOI:10.1145/3472456.3473524 The drug discovery process currently employed in the pharmaceutical industry typically requires about 10 years and $2–3 billion to deliver one new drug. This is both too expensive and too slow, especially in emergencies like the COVID-19 pandemic. In silico methodologies need to be improved both to select better lead compounds, so as to improve the efficiency of later stages in the drug discovery protocol, and to identify those lead compounds more quickly. No known methodological approach can deliver this combination of higher quality and speed. Here, we describe anIntegrated Modeling PipEline for COVID Cure by Assessing Better LEads (IMPECCABLE) that employs multiple methodological innovations to overcome this fundamental limitation.
Scalable HPC & AI Infrastructure for COVID-19 Therapeutics H. Lee, A. Merzky, L. Tan, M. Titov, M. Turilli, D. Alfe, A. Bhati, A. Brace, A. Clyde, P. V. Coveney, H. Ma, A. Ramanathan, R. Stevens, A. Trifan, H. Van Dam, S. Wan, S. Wilkinson, S. Jha, Platform for Advanced Scientific Computing Conference (PASC ’21), July 5-9 (2021), DOI: DOI:10.1145/3468267.3470573 COVID-19 has claimed more than 2.7×10^6 lives and resulted in over 124×10^6 infections. There is an urgent need to identify drugs that can inhibit SARS-CoV-2. We discuss innovations in computational infrastructure and methods that are accelerating and advancing drug design. Specifically, we describe several methods that integrate artificial intelligence and simulation-based approaches, and the design of computational infrastructure to support these methods at scale. We discuss their implementation, characterize their performance, and highlight science advances that these capabilities have enabled.
Ensembles Are Required to Handle Aleatoric and Parametric Uncertainty in Molecular Dynamics Simulation M. Vassaux, S. Wan, W. Edeling and P. V. Coveney, . Chem. Theory. Comput., 17, 5187 (2021), DOI: DOI:10.1021/acs.jctc.1c00526 Classical molecular dynamics is a computer simulation technique that is in widespread use across many areas of science, from physics and chemistry to materials, biology, and medicine. The method continues to attract criticism due its oft-reported lack of reproducibility which is in part due to a failure to submit it to reliable uncertainty quantification (UQ). Here we show that the uncertainty arises from a combination of (i) the input parameters and (ii) the intrinsic stochasticity of the method controlled by the random seeds. We find that robust statistical measures of uncertainty in molecular dynamics simulation require the use of ensembles in all contexts.
The effect of protein mutations on drug binding suggests ensuing personalised drug selection S. Wan, D. Kumar, V. Ilyin, U. Al Homsi, G. Sher, A. Knuth, P. V. Coveney, Sci. Rep., 11, 13452 (2021), DOI: 10.1038/s41598-021-92785-w The advent of personalised medicine promises a deeper understanding of mechanisms and therefore therapies. However, the connection between genomic sequences and clinical treatments is often unclear. We studied 50 breast cancer patients belonging to a population-cohort in the state of Qatar. From Sanger sequencing, we identified several new deleterious mutations in the estrogen receptor 1 gene (ESR1). The effect of these mutations on drug treatment in the protein target encoded by ESR1, namely the estrogen receptor, was achieved via rapid and accurate protein–ligand binding affinity interaction studies which were performed for the selected drugs and the natural ligand estrogen. Four nonsynonymous mutations in the ligand-binding domain were subjected to molecular dynamics simulation using absolute and relative binding free energy methods, leading to the ranking of the efficacy of six selected drugs for patients with the mutations. Our study shows that a personalised clinical decision system can be created by integrating an individual patient’s genomic data at the molecular level within a computational pipeline which ranks the efficacy of binding of particular drugs to variant proteins.
In-silico human electro-mechanical ventricular modelling and simulation for drug-induced pro-arrhythmia and inotropic risk assessment F. Margara, Z. J. Wanga, F. Levrero-Florencio, A. Santiago, M. Vázquez, A. Bueno-Orovioa and B. Rodriguez, Progress in Biophysics and Molecular Biology, 159, 58-74 (2021), DOI: 10.1016/j.pbiomolbio.2020.06.007 An investigation of computer models of the heart calibrated against experimental data from human ventricular electrophysiology. The models focus on the effect of electro-mechanical coupling and pharmacological action, opening new avenues for investigations of the complex interplay between the electrical and mechanical cardiac substrates, its modulation by pharmacological action, and its translation to tissue and organ models of cardiac pathology.
When We Can Trust Computers (and When We Can’t) P. V. Coveney and R. R. HighfieldPhil. Trans. R. Soc. A., 379, 20200409 (2021), DOI: 10.1098/rsta.2020.0067 With the unprecedented rise of computer power, there is a widespread expectation that digital computers could solve any problem. We explore their limits in the domains of science and engineering for simpler as well as more complex systems and discuss where analogue computers stand.
Finite element analysis informed variable selection for femoral fracture risk prediction M. Taylor, M. Viceconti, P. Bhattacharya, X. Li, Journal of the Mechanical Behavior of Biomedical Materials, 118, 104434 (2021) DOI: 10.1016/j.jmbbm.2021.104434 Logistic regression classification (LRC) is widely used to develop models to predict the risk of femoral fracture. LRC models based on areal bone mineral density (aBMD) alone are poor, with area under the receiver operator curve (AUROC) scores reported to be as low as 0.63. This has led to researchers investigating methods to extract further information from the image to increase performance. This raises the question, are we reaching the limits of the information that can be extracted from an image? Finite element analysis was used in combination with active shape and appearance modelling to select variables to develop LRC models of fracture risk. Based on the findings in this paper, it is suggested that we are reaching the limits of the information that can be extracted from an image to predict fracture risk.
Evaluation of patient tissue selection methods for deriving equivalent density calibration for femoral bone quantitative CT analyses C.Winsor, X. Lib, M. Qasim, C. R. Henaka, P. J. Pickhardt, H. Ploeg, M. Viceconti, Bone, 143 (2021) DOI: 10.1016/j.bone.2020.115759 An investigation of computer models of the heart calibrated against experimental data from human ventricular electrophysiology. The models focus on the effect of electro-mechanical coupling and pharmacological action, opening new avenues for investigations of the complex interplay between the electrical and mechanical cardiac substrates, its modulation by pharmacological action, and its translation to tissue and organ models of cardiac pathology.
TorchMD: A Deep Learning Framework for Molecular Simulations S. Doerr, M. Majewski, A. Pérez, A. Krämer, C. Clementi, F. Noe, T. Giorgino, and G. De Fabritiis, J. Chem. Theory Comput., 17, 4, 2355–2363 (2021), DOI: 10.1021/acs.jctc.0c01343 Molecular dynamics simulations provide a mechanistic description of molecules by relying on empirical potentials. The quality and transferability of such potentials can be improved leveraging data-driven models derived with machine learning approaches. Here, we present TorchMD, a framework for molecular simulations with mixed classical and machine learning potentials. All force computations including bond, angle, dihedral, Lennard-Jones, and Coulomb interactions are expressed as PyTorch arrays and operations. Moreover, TorchMD enables learning and simulating neural network potentials. We validate it using standard Amber all-atom simulations, learning an ab initio potential, performing an end-to-end training, and finally learning and simulating a coarse-grained model for protein folding. We believe that TorchMD provides a useful tool set to support molecular simulations of machine learning potentials.
Uncertainty Quantification in Classical Molecular Dynamics S. Wan, R. C. Sinclair and P. V. Coveney, Phil. Trans. R. Soc. A, 379, 20200082 (2021), DOI: 10.1098/rsta.2020.0082 Molecular dynamics simulation is now a widespread approach for understanding complex systems on the atomistic scale. It finds applications from physics and chemistry to engineering, life and medical science. In the last decade, the approach has begun to advance from being a computer-based means of rationalizing experimental observations to producing apparently credible predictions for a number of real-world applications within industrial sectors such as advanced materials and drug discovery. However, key aspects concerning the reproducibility of the method have not kept pace with the speed of its uptake in the scientific community. Here, we present a discussion of uncertainty quantification for molecular dynamics simulation designed to endow the method with better error estimates that will enable it to be used to report actionable results.
The influence of external electric fields on proton transfer tautomerism in the guanine-cytosine base pair A. Gheorghiu, P. V. Coveney and A. A. Arabi, Phys. Chem. Chem. Phys. 23, 6252-6265 (2021), DOI: 10.1039/D0CP06218A The Watson–Crick base pair proton transfer tautomers would be widely considered as a source of spontaneous mutations in DNA replication if not for their short lifetimes and thermodynamic instability. This work investigates the effects external electric fields have on the stability of the guanine–cytosine proton transfer tautomers within a realistic strand of aqueous DNA using a combination of ensemble-based classical molecular dynamics (MD) coupled to quantum mechanics/molecular mechanics (QM/MM).
The Impact of Uncertainty on Predictions of the CovidSim Epidemiological Code W. Edeling, H. Arabnejad, R. Sinclair, D. Suleimenova, K. Gopalakrishnan, B. Bosak, D. Groen, I. Mahmood, D. Crommelin and P. Coveney, Nat Comput Sci, 1, 128–135 (2021), DOI: 10.1038/s43588-021-00028-9 The MRC Centre for Global Infectious Disease Analysis at Imperial College London developed the CovidSim code, which was used to inform the UK Government’s response to the COVID-19 pandemic earlier in the year. We review the publicly available CovidSim epidemiological code by means of a parametric sensitivity analysis and uncertainty quantification and conclude that the model contains a large degree of uncertainty in its predictions, due to its inherent nature.
TIES 20: Relative Binding Free Energy with a Flexible Superimposition Algorithm and Partial Ring Morphing M. Bieniek, A. Bhati, S. Wan and P. V. Coveney, J. Chem. Theory Comput., 17, 2, 1250–1265 (2021), DOI: 10.1021/acs.jctc.0c01179 Thermodynamic integration with enhanced sampling (TIES) is a formally exact alchemical approach to the calculation of relative binding free energies. We implement a new, flexible-topology superimposition algorithm which improves the precision of the predicted free energies with respect to experimental data.
Pharmaceutical Industry—Academia Cooperation A. Heifetz, P. V. Coveney, D. G. Fedorov, I. Morao, T. James, M. Southey, K. Papadopoulos, M. J. Bodkin, A. Townsend-Nicholson, in: Mochizuki Y., Tanaka S., Fukuzawa K. (eds) Recent Advances of the Fragment Molecular Orbital Method, Springer, Singapore (2021), DOI: 10.1007/978-981-15-9235-5_15 We look at the long history of fruitful cooperation between academia and the pharmaceutical industry, the benefits and challenges for each, and provide some practical solutions, based on our own experiences and specific examples, to make this kind of collaboration successful and rewarding.

2020

Title Citation Summary
Electrophysiological and anatomical factors determine arrhythmic risk in acute myocardial ischaemia and its modulation by sodium current availability Hector Martinez-Navarro; Xin Zhou; Alfonso Bueno-Orovio; Blanca Rodriguez, Interface Focus (2020) DOI: 10.1098/rsfs.2019.0124 Investigating arrhythmia mechanisms in acute myocardial ischemia, the study utilizes over 300 high-performance computing simulations to explore the impact of variable sodium current availability. Despite sodium current blockers being used to mitigate arrhythmias, mutations leading to reduced sodium current availability increase arrhythmic risk in ischemic patients. By employing a human anatomically based torso-biventricular electrophysiological model, the research incorporates realistic ventricular anatomy and fiber architecture. Simulations reveal that decreased sodium current availability heightens arrhythmic risk due to both electrophysiological factors, such as increased refractoriness dispersion across the ischemic border zone, and anatomical factors, including conduction block from the thin right ventricle to the thick left ventricle. Notably, asymmetric ventricular anatomy amplifies arrhythmic risk for ectopic stimuli from the right ventricle and ventricular base. Interestingly, increasing sodium current availability proves ineffective in reducing arrhythmic risk for septo-basal ectopic excitation. Through human-based multiscale modeling, the study identifies crucial electrophysiological and anatomical factors determining arrhythmic risk in acute ischemia with variable sodium current availability.
Digital Blood in Massively Parallel CPU/GPU Systems for the Study of Platelet Transport: Supplementary Material Kotsalos, Christos; Latt, Jonas; Beny, Joel; Chopard, Bastien, Figshare (2020) DOI: 10.6084/m9.figshare.13213333 We introduce a flexible computational framework for simulating cellular blood flow, prioritizing high performance while maintaining accuracy and complexity. Our tool integrates Palabos for plasma simulation, a novel finite element method (FEM) solver for deformable blood cells, and an immersed boundary method for phase coupling. It supports hybrid CPU-GPU execution and modular component replacement. Our FEM-based solid dynamics kernel surpasses other FEM solvers and rivals mass-spring systems. Performance tests on Piz Daint confirm its efficiency, with case studies on platelet transport validating accuracy. This versatile framework balances accuracy and speed, ideal for future exascale architectures.
Parallel Multiphysics Coupling: Algorithmic and Computational Performances Guillaume Houzeaux; Marta Garcia-Gasulla; J.C. Cajas; R. Borrell; Alfonso Santiago; Charles Moulinec; Mariano Vázquez, International Journal of Computational Fluid Dynamics (2020) DOI: 10.1080/10618562.2020.1783440 This study investigates the performance of partitioned methods for solving multiphysics problems, which involve coupling different sets of partial differential equations. Partitioned methods handle each set of equations individually, iterating until convergence to obtain the monolithic solution. However, they incur an additional iterative loop, which can be implemented in parallel (a la Jacobi) or sequentially (a la Gauss-Seidel). Although the Gauss-Seidel method has inferior algorithmic properties compared to the Jacobi method, it utilizes computational resources more effectively. The research aims to evaluate the algorithmic and computational efficiencies of both coupling methods, focusing on multiphysics surface coupling. To improve the computational efficiency of the Gauss-Seidel method, an overloading strategy and an MPI barrier using the DLB library will be introduced. This approach aims to make the Gauss-Seidel method nearly as parallel efficient as the Jacobi method. The methodology relies on simple performance models and the solution of multiphysics problems to validate the proposed approach. By assessing both algorithmic and computational performance, the study aims to provide insights into optimizing partitioned methods for solving multiphysics surface coupling problems.
Supplemental Figures and Table from How quickly can we predict trimethoprim resistance using alchemical free energy methods? Fowler, Philip W, Figshare (2020) DOI: 10.6084/m9.figshare.12979029.v1 Whole-genome sequencing offers a promising avenue for improving clinical microbiology diagnostics, particularly for antibiotic-resistant infections. However, it struggles with rare or novel genetic variants, necessitating predictive methods. In Staphylococcus aureus, non-synonymous mutations in the dfrB gene can predict trimethoprim resistance by assessing their impact on binding free energy. Utilizing alchemical free energy methods, we demonstrate that 15 binding free energies calculated from short molecular dynamics simulations effectively classify all seven mutations in a clinical test set. With results obtainable in under an hour, this approach showcases the speed and reliability required for clinical implementation.
Supporting Material from Haemodynamic flow conditions at the initiation of high-shear platelet aggregation: a combined in vitro and cellular in silico study Rooij, B.J.M. Van; Závodszky, G.; Hoekstra, A.G.; Ku, D.N., Figshare (2020) DOI: 10.6084/m9.figshare.13272993.v1 This study aims to explore the impact of high shear rates on initial platelet aggregation in high-shear thrombosis. Using cell-based blood flow simulations in a microfluidic device, haemodynamic conditions are investigated and compared with in vitro platelet aggregation experiments using porcine whole blood (WB) and platelet-rich-plasma (PRP). Results reveal elevated platelet fluxes at the wall in silico, with WB exhibiting double the flux of PRP. Initial platelet aggregation and occlusion occur in vitro, particularly in the stenotic region, with PRP forming occlusive thrombi further downstream than WB. Comparison of shear rates and stresses between cell-based and continuum simulations suggests the latter is a suitable approximation for PRP but not for WB, particularly near the wall.
All-Optical Electrophysiology Refines Populations of In Silico Human iPSC-CMs for Drug Evaluation Michelangelo Paci, Elisa Passini, Aleksandra Klimas, Stefano Severi, Jari Hyttinen, Blanca Rodriguez, and Emilia Entcheva, Biophysical Journal (2020) DOI: 10.1016/j.bpj.2020.03.018 Recent advancements in high-throughput in vitro drug assays, fueled by human induced pluripotent stem cell-derived cardiomyocyte (hiPSC-CM) technology and contact-free all-optical systems, have revolutionized the simultaneous measurement of action potentials (APs) and Ca2+ transients (CaTrs). Complementing these experimental breakthroughs, parallel computational progress in in silico simulations has enabled accurate prediction of drug effects. This study combines these technologies, utilizing high-throughput experimental data to refine in silico hiPSC-CM populations and predict drug action mechanisms. By integrating optically obtained hiPSC-CM APs and CaTrs into in silico models, the study successfully predicts drug effects and offers mechanistic insights, emphasizing the synergy between high-throughput in vitro and population in silico approaches.
Blinded In Silico Drug Trial Reveals the Minimum Set of Ion Channels for Torsades de Pointes Risk Assessment Zhou, X., Qu, Y., Passini, E., Bueno-Orovio, A., Liu, Y., Vargas, H. M., & Rodriguez, B., Frontiers in Pharmacology (2020) DOI: 10.3389/fphar.2019.01643 Torsades de Pointes (TdP), a ventricular arrhythmia, poses a significant challenge in drug safety evaluation due to its potential drug-induced occurrence. Previous evaluations mainly focused on hERG channels, leading to incomplete predictions. To address this, the Comprehensive in Vitro Proarrhythmia Assay suggests integrating multiple ion channel effects. This study investigates the minimum necessary ion channels for reliable TdP risk prediction and the impact of variable IC50 and Hill coefficient values on prediction outcomes. Using a computational model optimized for repolarization abnormalities, results confirm that Nav1.5 (peak), Cav1.2, and hERG are essential for reliable predictions, and moderate variations in IC50 and Hill coefficients can affect accuracy.
Computational prediction of GPCR oligomerization A. Townsend-Nicholson, N. Altwaijry, A. Potterton, I. Morao, A. Heifetz, Current Opinion in Structural Biology (2020) DOI: 10.1016/j.sbi.2019.04.005 Recent advances in G protein-coupled receptor (GPCR) research have led to a surge in solved crystal structures, shedding light on active and inactive forms and their interactions with various ligands and proteins. However, understanding the precise configuration of GPCR oligomers in different biologically relevant states remains challenging due to limited experimental data. Computational methods offer a promising avenue to predict the formation of receptor dimers and higher order oligomers. Ensemble-based computational approaches, combined with quantum mechanical analyses of molecular interactions, can provide reproducible predictions of GPCR dimerization. These predictions can inform future research aimed at uncovering previously unidentified GPCR dimers and expanding the repertoire of experimentally determined oligomeric GPCR structures.
Conformational Searching with Quantum Mechanics Matthew Habgood, Tim James, Alexander Heifetz, Quantum Mechanics in Drug Discovery (2020) DOI: 10.1007/978-1-0716-0282-9_14 Conformational searching is vital in computer-aided drug design, with quantum mechanical (QM) simulations offering enhanced accuracy but at a higher computational expense compared to forcefield methods. This chapter explores the application of QM to conformational searching, addressing challenges such as generating ensembles approximating a molecule’s bioactive conformation, analyzing and modifying bioactive conformations, and approximating solution-phase ensembles alongside NMR data. While the impact of QM on high-throughput applications is debatable, it proves valuable for lower-throughput tasks. Choosing the optimal QM method is also discussed, with DFT methods requiring large basis sets for acceptable results, yet some papers demonstrate useful outcomes with more cost-effective methods, presenting a dilemma yet to be resolved in the literature.
Dual Transcriptomic and Molecular Machine Learning Predicts all Major Clinical Forms of Drug Cardiotoxicity Polina Mamoshina, Alfonso Bueno-Orovio, and Blanca Rodriguez, Frontiers in Pharmacology (2020) DOI: 10.3389/fphar.2020.00639 Computational methods are crucial for enhancing drug discovery pipelines, particularly in identifying cardiotoxicity. Leveraging machine learning on a vast dataset of transcriptional and molecular profiles, we successfully predict and preserve relationships for six types of drug-induced cardiotoxicity. Validation on an independent dataset and cross-validation demonstrate the algorithm’s robustness, achieving an average accuracy of 79% in discriminating safe versus risky drugs and accurately predicting individual cardiotoxicities. Independent testing on additional datasets yields similar results, suggesting the methodology’s potential applicability to various tissue-specific side effects beyond cardiotoxicity.
Exascale potholes for HPC: Execution performance and variability analysis of the flagship application code HemeLB B. J. N. Wylie, IEEE/ACM International Workshop on HPC User Support Tools (HUST) and Workshop on Programming and Performance Visualization Tools (ProTools) (2020) DOI: 10.1109/HUSTProtools51951.2020.00014 Performance measurement and analysis of parallel applications face challenges, especially with the advent of exascale computing. This study examines the performance of the HemeLB application code on the SuperMUC-NG system using the Performance Optimisation and Productivity (POP) CoE methodology. Despite maintaining 80% scaling efficiency with over 100,000 MPI processes, initial performance issues were traced to faulty compute nodes, impacting strong scaling. Excluding these nodes improved performance significantly, achieving a 190x speed-up compared to fewer MPI processes. Communication efficiency remains high, but load balance limitations, attributed to core-to-core variability and memory access stalls, affect parallel efficiency. The study showcases the POP methodology’s effectiveness in diagnosing and addressing performance issues.
From digital hype to analogue reality: Universal simulation beyond the quantum and exascale eras Peter V. Coveney, Roger R. Highfield, Journal of Computational Science (2020) DOI: 10.1016/j.jocs.2020.101093 Innovation’s future, many argue, hinges on simulation. With advancing computing power, discussions surrounding their potential span numerous domains: from subatomic physics to economics. The impending quantum and exascale computing era sees machine learning and artificial intelligence significantly influencing the field. This article traces simulation’s history, underscores the importance of deeper mechanistic understanding for potent machine learning, and explores the potential of exascale and quantum computing. It also delineates the limitations of digital computing, classical and quantum alike, while advocating for a discerning approach to assessing the future of modelling and simulation, wherein analogue computing is poised to play a pivotal role.
Human Purkinje in silico model enables mechanistic investigations into automaticity and pro-arrhythmic abnormalities. Trovato, C; Passini, E; Nagy, N; Varro, A; Abi-Gerges, N; Severi, S; Rodriguez, B, Journal of Molecular and Cellular Cardiology (2020) DOI: 10.1016/j.yjmcc.2020.04.001 This study presents a novel human Purkinje cell (PCs) electrophysiology computational model, Trovato2020, aimed at elucidating ionic mechanisms underlying Purkinje-related electrophysiology and arrhythmogenicity. Developed, calibrated, and validated using human experimental data, the model incorporates detailed Purkinje-specific ionic currents and Ca2+ handling. Simulations demonstrate that the Trovato2020 model reproduces key action potential (AP) features consistent with human Purkinje recordings, including automaticity, early afterdepolarizations (EADs), and delayed afterdepolarizations (DADs). Furthermore, multiscale investigations using an experimentally-calibrated population of PCs reveal insights into the mechanisms underlying EADs and DADs, shedding light on the role of cardiac Purkinje cells in ventricular arrhythmias.
Lattice-Boltzmann interactive blood flow simulatSensitivity analysis of a strongly-coupled human-based electromechanical cardiac model: Effect of mechanical parameters on physiologically relevant biomarkersion pipeline S. S. Esfahani, X. Zhai, M. Chen, A. Amira, F. Bensaali, J. AbiNahed, S. Dakua, G. Younes, A. Baobeid, R. A. Richardson and P. V. Coveney, International Journal of Computer Assisted Radiology and Surgery (2020) DOI: 10.1007/s11548-020-02120-3 Cerebral aneurysms pose significant clinical challenges, with interventional radiology treatment reliant on radiologist expertise. This study introduces a pipeline for cerebral blood flow simulation and real-time visualization, addressing this need. Using an enhanced version of HemeLB as the computational core and a CUDA-based ray marching method for visualization, the pipeline achieves superior scalability and update rates compared to previous methods. Results show over two-fold speed improvement, enabling real-time 3D visualization. This reliable modeling and visualization environment offers insights into cerebral aneurysm hemodynamics, enhancing clinical decision-making with real-time blood flow information for clinicians.
The ‘Digital Twin’ to enable the vision of precision cardiology Corral-Acero J, Margara F, Marciniak M, Rodero C, Loncaric F, Feng Y, Gilbert A, Fernandes JF, Bukhari HA, Wajdan A, Martinez MV, Santos MS, Shamohammdi M, Luo H, Westphal P, Leeson P, DiAchille P, Gurev V, Mayr M, Geris L, Pathmanathan P, Morrison T, Cornelussen R, Prinzen F, Delhaas T, Doltra A, Sitges M, Vigmond EJ, Zacur E, Grau V, Rodriguez B, Remme EW, Niederer S, Mortier P, McLeod K, Potse M, Pueyo E, Bueno-Orovio A, Lamata P., European Heart Journal (2020) DOI: 10.1093/eurheartj/ehaa159 Precision medicine aims to tailor therapies to individual patients, leveraging extensive patient data. The second pillar enabling this vision is the advancing power of computers and algorithms to create a patient’s ‘digital twin’. Computational models enhance diagnostic and prognostic capabilities, projecting pathways for health restoration based on model predictions. This paper reviews early steps in cardiovascular medicine’s digital twin development, highlighting challenges and opportunities. It emphasizes the synergy between mechanistic and statistical models, accelerating cardiovascular research and advancing precision medicine’s goals.
Guiding Medicinal Chemistry with Fragment Molecular Orbital (FMO) Method A. Heifetz, T. James, M. Southey, M. J. Bodkin, S. Bromidge, Quantum Mechanics in Drug Discovery. Methods in Molecular Biology, vol. 2114, 37-48 (2020) DOI: 10.1007/978-1-0716-0282-9_3 The understanding of binding interactions between a protein and a small molecule plays a key role in the rationalization of potency and selectivity and in design of new ideas. However, even when a target of interest is structurally enabled, visual inspection and force field-based molecular mechanics calculations cannot always explain the full complexity of the molecular interactions that are critical in drug design. In this chapter, we describe the FMO method and illustrate its application in the discovery of the benzothiazole (BZT) series as novel tyrosine kinase ITK inhibitors for treatment of allergic asthma.
Accurate Scoring in Seconds with the Fragment Molecular Orbital and Density-Functional Tight-Binding Methods I. Morao, A. Heifetz, D. G. Fedorov, Quantum Mechanics in Drug Discovery. Methods in Molecular Biology, vol. 2114, 143-148 (2020) DOI: 10.1007/978-1-0716-0282-9_9 The accurate evaluation of receptor-ligand interactions is an essential part of rational drug design. While quantum mechanical (QM) methods have been a promising means by which to achieve this, traditional QM is not applicable for large biological systems due to its high computational cost. Here, the fragment molecular orbital (FMO) method has been combined with the density-functional tight-binding (DFTB) method to compute energy calculations of biological systems in seconds. For the first time, it is now possible to perform FMO calculations in a high-throughput manner.
Femoral neck strain prediction during level walking using a combined musculoskeletal and finite element model approach Z. Altai, E. Montefiori, B. van Veen, M. A. Paggiosi, E. V. McCloskey, M. Viceconti, C. Mazzà, X. Li, PLoS ONE, 16(2):
e0245121 (2020) DOI: 10.1371/journal.pone.0245121
Recently, coupled musculoskeletal-finite element modelling approaches have emerged as a way to investigate femoral neck loading during various daily activities. Combining personalised gait data with finite element models will not only allow us to study changes in motion/movement, but also their effects on critical internal structures, such as the femur. However, previous studies have been hampered by the small sample size and the lack of fully personalised data in order to construct the coupled model. Therefore, the aim of this study was to build a pipeline for a fully personalised multiscale (body-organ level) model to investigate the strain levels at the femoral neck during a normal gait cycle. Five postmenopausal women were included in this study. The current findings suggest that personal variations are substantial, and hence it is important to consider multiple subjects before deriving general conclusions for a target population.
Biorheology of occlusive thrombi formation under high shear: in vitro growth and shrinkage B. J. M. van Rooij, G. Závodszky, A. G. Hoekstra and D. N. Ku, Scientific Reports, 10, 18604 (2020) DOI: 10.1038/s41598-020-74518-7 Occlusive thrombi formed under high flow shear rates develop very rapidly in arteries and may lead to myocardial infarction or stroke. Rapid platelet accumulation (RPA) and occlusion of platelet-rich thrombi and clot shrinkage have been studied after flow arrest. However, the influence of margination and shear rate on occlusive clot formation is not fully understood yet. In this study, the influence of flow on the growth and shrinkage of a clot is investigated.
A Heterogeneous Multi-scale Model for Blood Flow B. Czaja, G. Závodszky, A. Hoekstra, In: Computational Science – ICCS 2020, part of ICCS 2020 Lecture Notes in Computer Science, vol. 12142, 403-409, (2020), DOI: 10.1007/978-3-030-50433-5_31 This research focuses on developing a heterogeneous multi-scale model (HMM) for blood flow. Two separate scales are considered in this study, a Macro-scale, which models whole blood as a continuous fluid and tracks the transport of hematocrit profiles through an advection diffusion solver. And a Micro-scale, which computes directly local diffusion coefficients and viscosities using cell resolved simulations. The coupling between these two scales also includes the use of a surrogate model, which saved local viscosity and diffusion coefficients from previously simulated local hematocrit and shear rate combinations. As the HMM model progresses fewer micro models will be spawned. This is accomplished through the surrogate by interpolating from previously computed viscosities and diffusion coefficients.
Characterizing Protein-Protein Interactions with the Fragment Molecular Orbital Method A. Heifetz, V. Sladek, A. Townsend-Nicholson, D. G. Fedorov, In: Quantum Mechanics in Drug Discovery, part of Methods in Molecular Biology, vol. 2114, 187-205 (2020), DOI: 10.1007/978-1-0716-0282-9_13 Proteins are vital components of living systems, serving as building blocks, molecular machines, enzymes, receptors, ion channels, sensors, and transporters. Protein-protein interactions (PPIs) are a key part of their function. In this chapter, we have demonstrated how three different FMO-based approaches (pair interaction energy analysis (PIE analysis), subsystem analysis (SA) and analysis of protein residue networks (PRNs)) have been applied to study PPI in three protein-protein complexes.
Characterizing Rhodopsin-Arrestin Interactions with the Fragment Molecular Orbital (FMO) Method A. Heifetz, A. Townsend-Nicholson, In: Quantum Mechanics in Drug Discovery, part of Methods in Molecular Biology, vol. 2114, 177-186 (2020), DOI: 10.1007/978-1-0716-0282-9_12 Arrestin binding to G protein-coupled receptors (GPCRs) plays a vital role in receptor signaling. Recently, the crystal structure of rhodopsin bound to activated visual arrestin was resolved using XFEL (X-ray free electron laser). However, even with the crystal structure in hand, our ability to understand GPCR-arrestin binding is limited by the availability of accurate tools to explore receptor-arrestin interactions.
Analyzing GPCR-Ligand Interactions with the Fragment Molecular Orbital (FMO) Method A. Heifetz, T. James, M. Southey, I. Morao, D. G. Fedorov, M. J. Bodkin, A. Townsend-Nicholson, In: Quantum Mechanics in Drug Discovery, part of Methods in Molecular Biology, vol. 2114, 163-175 (2020), DOI: 10.1007/978-1-0716-0282-9_11 G-protein-coupled receptors (GPCRs) have enormous physiological and biomedical importance, and therefore it is not surprising that they are the targets of many prescribed drugs. Further progress in GPCR drug discovery is highly dependent on the availability of protein structural information. In this chapter, we describe how to use FMO in the characterization of GPCR-ligand interactions.
Characterizing Interhelical Interactions of G-Protein Coupled Receptors with the Fragment Molecular Orbital Method A. Heifetz, I. Morao, M. Madan Babu, T. James, M. W. Y. Southey, D. G. Fedorov, M. Aldeghi, M. J. Bodkin, and A. Townsend-Nicholson, J. Chem. Theory Comput. 16, 4, 2814–2824 (2020), DOI: 10.1021/acs.jctc.9b01136 G-protein coupled receptors (GPCRs) are the largest superfamily of membrane proteins, regulating almost every aspect of cellular activity and serving as key targets for drug discovery. We have identified an accurate and reliable computational method to characterize the strength and chemical nature of the interhelical interactions between the residues of transmembrane (TM) domains during different receptor activation states, something that cannot be characterized solely by visual inspection of structural information.
Spherization of red blood cells and platelet margination in COPD patients K. Boudjeltia, C. Kotsalos, D. de Sousa, A. Rousseau C. Lelubre, O. Sartenaer, M. Piagnerelli, J. Dohet‐Eraly, F. Dubois, N. Tasiaux, B. Chopard, A. Van Meerhaeghe, Annals Reports 1485, 1, 71-82 (2020), DOI: 10.1111/nyas.14489 Red blood cells (RBCs) in pathological situations undergo biochemical and conformational changes, leading to alterations in rheology involved in cardiovascular events. The shape of RBCs in volunteers and stable and exacerbated chronic obstructive pulmonary disease (COPD) patients was analyzed. The effects of RBC spherization on platelet transport (displacement in the flow field caused by their interaction with RBCs) were studied in vitro and by numerical simulations. RBC spherization was observed in COPD patients compared with volunteers.
Small Molecule Modulation of Intrinsically Disordered Proteins Using Molecular Dynamics Simulations P. Herrera-Nieto, A. Pérez, and G. De Fabritiis, J. Chem. Inf. Model. 60, 10, 5003–5010 (2020), DOI: 10.1021/acs.jcim.0c00381 The extreme dynamic behavior of intrinsically disordered proteins hinders the development of drug-like compounds capable of modulating them. There are several examples of small molecules that specifically interact with disordered peptides. However, their mechanisms of action are still not well understood. Here, we use extensive molecular dynamics simulations combined with adaptive sampling algorithms to perform free ligand binding studies in the context of intrinsically disordered proteins. The results show several protein–ligand bound states characterized by the establishment of a loosely oriented interaction mediated by a limited number of contacts between the ligand and critical residues of p27.
Characterization of partially ordered states in the intrinsically disordered N-terminal domain of p53 using millisecond molecular dynamics simulations P. Herrera-Nieto, A. Pérez & G. De Fabritiis, Scientific Reports 10, 12402 (2020) DOI: 10.1038/s41598-020-69322-2 The exploration of intrinsically disordered proteins in isolation is a crucial step to understand their complex dynamical behavior. In particular, the emergence of partially ordered states has not been explored in depth. The experimental characterization of such partially ordered states remains elusive due to their transient nature. Molecular dynamics mitigates this limitation thanks to its capability to explore biologically relevant timescales while retaining atomistic resolution. Here, millisecond unbiased molecular dynamics simulations were performed in the exemplar N-terminal region of p53. Our research highlights the potential complexity underlying the energy surface of intrinsically disordered proteins.
Hydrodynamic Resistance of Intracranial Flow-Diverter Stents: Measurement Description and Data Evaluation B. Csippa, D. Gyürki, G. Závodszky, I. Szikora and G. Paál, Cardiovascular Engineering and Technology 11, 1–13 (2020) DOI: 10.1007/s13239-019-00445-y Intracranial aneurysms are malformations forming bulges on the walls of brain arteries. A flow diverter device is a fine braided wire structure used for the endovascular treatment of brain aneurysms. This work presents a rig and a protocol for the measurement of the hydrodynamic resistance of flow diverter stents. Hydrodynamic resistance is interpreted here as the pressure loss versus volumetric flow rate function through the mesh structure. The difficulty of the measurement is the very low flow rate range and the extreme sensitivity to contamination and disturbances. Based on our evaluation method a confidence band can be determined for a given deployment scenario. Collectively analysing the hydrodynamic resistance and the geometric indices, a deeper understanding of an implantation can be obtained. Our results suggest that to correctly interpret the hydrodynamic resistance of a scenario, the deployment length has to be considered.
A particle-based model for endothelial cell migration under flow conditions P. S. Zun, A. J. Narracott, P. C. Evans, B. J. M. van Rooij and A. G. Hoekstra, Biomechanics and Modeling in Mechanobiology 19, 681–692 (2020) DOI: 10.1007/s10237-019-01239-w Endothelial cells (ECs) play a major role in the healing process following angioplasty to inhibit excessive neointima. This makes the process of EC healing after injury, in particular EC migration in a stented vessel, important for recovery of normal vessel function. In that context, we present a novel particle-based model of EC migration and validate it against in vitro experimental data. The results of this study support the hypothesis that EC migration is strongly affected by the direction and magnitude of local wall shear stress.
Investigating rolling as mechanism for humeral fractures in non-ambulant infants: a preliminary finite element study Z. Altai, M. Viceconti, X. Li, A. C. Offiah, Clinical Radiology 75, 1, 78 (2020) DOI: 10.1016/j.crad.2019.08.026 This study aims to use personalised computed tomography (CT)-based finite element models to quantitatively investigate the likelihood of self-inflicted humeral fracture in non-ambulant infants secondary to rolling. Results of this study challenge the plausibility of self-inflicted humeral fracture caused by rolling in non-ambulant infants and indicate that it is unlikely for a humeral fracture to result from this mechanism without the assistance of an external force.
AdaptiveBandit: A Multi-armed Bandit Framework for Adaptive Sampling in Molecular Simulations A. Pérez, P. Herrera-Nieto, S. Doerr, and G. De Fabritiis, J. Chem. Theory Comput. 16, 7, 4685–4693 (2020) DOI: 10.1021/acs.jctc.0c00205 Sampling from the equilibrium distribution has always been a major problem in molecular simulations due to the very high dimensionality of the conformational space. Over several decades, many approaches have been used to overcome the problem. In particular, we focus on unbiased simulation methods such as parallel and adaptive sampling. Here, we recast adaptive sampling schemes on the basis of multi-armed bandits and develop a novel adaptive sampling algorithm under this framework, AdaptiveBandit. We test it on multiple simplified potentials and in a protein folding scenario. We find that this framework performs similarly to or better than previous methods in every type of test potential. Furthermore, it provides a novel framework to develop new sampling algorithms with better asymptotic characteristics.
Coarse graining molecular dynamics with graph neural networks B. E. Husic, N. E. Charron, D. Lemm, J. Wang, A. Pérez, M. Majewski, A. Krämer, Y. Chen, S. Olsson, G. de Fabritiis, F. Noé and Cecilia Clementi, J. Chem. Phys. 153, 194101 (2020), DOI: 10.1063/5.0026133 Coarse graining enables the investigation of molecular dynamics for larger systems and at longer timescales than is possible at an atomic resolution. However, a coarse graining model must be formulated such that the conclusions we draw from it are consistent with the conclusions we would draw from a model at a finer level of detail. It has been proved that a force matching scheme defines a thermodynamically consistent coarse-grained model for an atomistic system in the variational limit. We introduce a hybrid architecture for the machine learning of coarse-grained force fields that learn their own features via a subnetwork that leverages continuous filter convolutions on a graph neural network architecture. We demonstrate that this framework succeeds at reproducing the thermodynamics for small biomolecular systems. Since the learned molecular representations are inherently transferable, the architecture presented here sets the stage for the development of machine-learned, coarse-grained force fields that are transferable across molecular systems.
SkeleDock: A Web Application for Scaffold Docking in PlayMolecule A. Varela-Rial, M. Majewski, A. Cuzzolin, G. Martínez-Rosell and Gianni De Fabritiis, J. Chem. Inf. Model. 60, 6, 2673–2677 (2020), DOI: 10.1021/acs.jcim.0c00143 SkeleDock is a scaffold docking algorithm which uses the structure of a protein–ligand complex as a template to model the binding mode of a chemically similar system. This algorithm was evaluated in the D3R Grand Challenge 4 pose prediction challenge, where it achieved competitive performance. Furthermore, we show that if crystallized fragments of the target ligand are available then SkeleDock can outperform rDock docking software at predicting the binding mode. This Application Note also addresses the capacity of this algorithm to model macrocycles and deal with scaffold hopping.
Large scale relative protein ligand binding affinities using non-equilibrium alchemy V. Gapsys, L. Pérez-Benito, M. Aldeghi, D. Seeliger, Herman van Vlijmen, G. Tresadern, and Bert L. de Groot, Chemical Science, 4, (2020), DOI: 10.1039/C9SC03754C Ligand binding affinity calculations based on molecular dynamics (MD) simulations and non-physical (alchemical) thermodynamic cycles have shown great promise for structure-based drug design. However, their broad uptake and impact is held back by the notoriously complex setup of the calculations. Only a few tools other than the free energy perturbation approach by Schrödinger Inc. (referred to as FEP+) currently enable end-to-end application. Here, we present for the first time an approach based on the open-source software pmx that allows to easily set up and run alchemical calculations for diverse sets of small molecules using the GROMACS MD engine. The method relies on theoretically rigorous non-equilibrium thermodynamic integration (TI) foundations, and its flexibility allows calculations with multiple force fields. In this study, results from the Amber and Charmm force fields were combined to yield a consensus outcome performing on par with the commercial FEP+ approach. For the first time, a setup is presented for overall high precision and high accuracy relative protein–ligand alchemical free energy calculations based on open-source software.
HPC compact quasi-Newton algorithm for interface problems A. Santiago, M. Zavala-Akéc, R. Borrella, G. Houzeaux, M. Vázquez, Journal of Fluids and Structures, 96, 103009 (2020), DOI: 10.1016/j.jfluidstructs.2020.103009 In this work we present a robust interface coupling algorithm called Compact Interface quasi-Newton (CIQN). It is designed for computationally intensive applications using an MPI multi-code partitioned scheme. The algorithm allows to reuse information from previous time steps, feature that has been previously proposed to accelerate convergence. Through algebraic manipulation, an efficient usage of the computational resources is achieved by: avoiding construction of dense matrices and reduce every multiplication to a matrix–vector product and reusing the computationally expensive loops. This leads to a compact version of the original quasi-Newton algorithm. The novelty of this article lies in the HPC focused implementation of the algorithm, detailing how to fuse and combine the composing blocks to obtain an scalable MPI implementation. Such an implementation is mandatory in large scale cases, for which the contact surface cannot be stored in a single computational node, or the number of contact nodes is not negligible compared with the size of the domain.
Fluid–structure interaction simulations outperform computational fluid dynamics in the description of thoracic aorta haemodynamics and in the differentiation of progressive dilation in Marfan syndrome patients R. Pons, A. Guala, J. F. Rodríguez-Palomares, J. C. Cajas, L. Dux-Santoy, G. Teixidó-Tura, J. J. Molins, M. Vázquez, A. Evangelista and J. Martorell, R. Soc. open sci., 7, 191752 (2020), DOI: 10.1098/rsos.191752 Abnormal fluid dynamics at the ascending aorta may be at the origin of aortic aneurysms. This study was aimed at comparing the performance of computational fluid dynamics (CFD) and fluid–structure interaction (FSI) simulations against four-dimensional (4D) flow magnetic resonance imaging (MRI) data; and to assess the capacity of advanced fluid dynamics markers to stratify aneurysm progression risk. Fluid dynamic simulations of the thoracic aorta require fluid–solid interaction to properly reproduce complex haemodynamics. FSI- but not CFD-derived SSR could help stratifying MFS patients.
Computational biomedicine. Part II: organs and systems – Introduction P. V. Coveney, A. Hoekstra, B. Rodriguez and M. Viceconti, J R Soc Interface Focus, 11, 20200082 (2020), DOI: 10.1098/rsfs.2020.0082 Introduction to the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Haemodynamic flow conditions at the initiation of high-shear platelet aggregation: a combined in vitro and cellular in silico study B. J. M. van Rooij, G. Závodszky, A. G. Hoekstra and D. N. Ku, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Digital blood in massively parallel CPU/GPU systems for the study of platelet transport C. Kotsalos, J. Latt, J. Beny and B. Chopard, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Coupling one-dimensional arterial blood flow to three-dimensional tissue perfusion models for in silico trials of acute ischaemic stroke R. M. Padmos, T. I. Józsa, W. K. El-Bouri, P. R. Konduri, S. J. Payne and A. G. Hoekstra, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
A porous circulation model of the human brain for in silico clinical trials in ischaemic stroke T. I. Józsa, R. M. Padmos, N. Samuels, W. K. El-Bouri, A. G. Hoekstra and S. J. Payne, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Applicability assessment of a stent-retriever thrombectomy finite-element model G. Luraghi, J. Felix Rodriguez Matas, G. Dubini, Francesca Berti, S. Bridio, S. Duffy, A. Dwivedi, R. McCarthy, B. Fereidoonnezhad, P. McGarry, C. B. L. M. Majoie, F. Migliavacca and on behalf of the INSIST investigators, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
The EurValve model execution environment M. Bubak, K. Czechowicz, T. Gubała, D. R. Hose, M. Kasztelnik, M. Malawski, J. Meizner, P. Nowakowski and S. Wood, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Towards blood flow in the virtual human: efficient self-coupling of HemeLB J. W. S. McCullough, R. A. Richardson, A. Patronis, R. Halver, R. Marshall, M. Ruefenacht, B. J. N. Wylie, T. Odaker, M. Wiedemann, B. Lloyd, E. Neufeld, G. Sutmann, A. Skjellum, D. Kranzlmüller and P. V. Coveney, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection. We report on the progress of the HemeLB lattice Boltzmann code in simulating 3D macroscopic blood flow on a full human scale. The work is in context of the grand aim to create a virtual human – a personalised, digital copy of an individual that will assist in a patient’s diagnosis, treatment and recovery. Integral to the construction of a virtual human, we outline the implementation of a self-coupling strategy for HemeLB.
Analysis of mechanotransduction dynamics during combined mechanical stimulation and modulation of the extracellular-regulated kinase cascade uncovers hidden information within the signalling noise G. Ascolani, T. M. Skerry, D. Lacroix, E. Dall’Ara and A. Shuaib, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Deep medical image analysis with representation learning and neuromorphic computing N. Getty, T. Brettin, D. Jin, R. Stevens and F. Xia, J R Soc Interface Focus, 11, 20200082 (2020), 10.1098/rsfs.2020.0082 Part of the Computational biomedicine. Part II: organs and systems Theme Issue collection.
Computational biomedicine. Part I: molecular medicine – Introduction S. Wan, A. Potterton, F. S. Husseini, D. W. Wright, A. Heifetz, M. Malawski, A. Townsend-Nicholson and P. V. Coveney, J R Soc Interface Focus, 10, 20190128 (2020), DOI: 10.1098/rsfs.2020.0047 Introduction to the Computational biomedicine. Part I: molecular medicine Theme Issue collection.
Hit-to-lead and lead optimization binding free energy calculations for G protein-coupled receptors S. Wan, A. Potterton, F. S. Husseini, D. W. Wright, A. Heifetz, M. Malawski, A. Townsend-Nicholson and P. V. Coveney, J R Soc Interface Focus, 10, 20190128 (2020), DOI: 10.1098/rsfs.2019.0128 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. This paper presents an application of the ESMACS and TIES methods to compute the binding free energies of a series of ligands at G protein-coupled receptors.
On the faithfulness of molecular mechanics representations of proteins towards quantum-mechanical energy surfaces G. König and S. Riniker, J R Soc Interface Focus, 10, 20190121 (2020), DOI: 10.1098/rsfs.2019.0121 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. In this paper, various classical force fields based on molecular mechanics are assessed against quantum mechanical predictions in order to conclude on their accuracy in modelling the relationship between protein structure and function.
How quickly can we predict trimethoprim resistance using alchemical free energy methods? P. W. Fowler, J R Soc Interface Focus, 10, 20190141 (2020), DOI: 10.1098/rsfs.2019.0141 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. This work discusses the efficiency of the alchemical free energy method in predicting antimicrobial resistance to the antibiotic trimethoprim, as an alternative method to genome sequencing of the pathogen.
Large-scale binding affinity calculations on commodity compute clouds S. J. Zasada, D. W. Wright and P. V. Coveney, J R Soc Interface Focus, 10, 20190133 (2020), DOI: 10.1098/rsfs.2019.0133 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. This paper presents an automated workflow for calculating calculate the binding affinities of compounds bound to proteins known as the binding affinity calculator (BAC). BAC automates the process of calculating free energies from the stage of initial model building, through ensemble averaging, to data analysis.
Rapid, accurate, precise and reproducible ligand–protein binding free energy prediction S. Wan, A. P. Bhati, S. J. Zasada and P. V. Coveney, J R Soc Interface Focus, 10, 20200007 (2020), DOI: 10.1098/rsfs.2020.0007 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. Predicting the binding affinity between molecules accurately and efficiently has posed major theoretical and computational challenges. We review a few methods -including two of our own- in terms of how they respond to those challenges and show how they can be used in real-world problems such as hit-to-lead and lead optimization stages in drug discovery, and in personalized medicine.
The influence of base pair tautomerism on single point mutations in aqueous DNA A. Gheorghiu, P. V. Coveney and A. A. Arabi, J R Soc Interface Focus, 10, 20190120 (2020), DOI: 10.1098/rsfs.2019.0120 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. Mutations within DNA are crucial to both natural evolution and the occurrence of genetic diseases, and are due to a number of different causes. One such cause is known as tautomerism. This paper investigated its kinetics and thermodynamics.
Quantum computing using continuous-time evolution V. Kendon, J R Soc Interface Focus, 10, 20190143 (2020), DOI: 10.1098/rsfs.2019.0143 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. As digital silicon computers are reaching their limits in terms of speed, other types of computation using radically different architectures, including neuromorphic and quantum, promise breakthroughs in both speed and efficiency. This article outlines the current state-of-the-art and future prospects for quantum computing, and provides some indications of how and where to apply it to speed up bottlenecks in biological simulation.
Educating and engaging new communities of practice with high performance computing through the integration of teaching and research A. Townsend-Nicholson, J R Soc Interface Focus, 10, 20200003 (2020), DOI: 10.1098/rsfs.2020.0003 Part of the Computational biomedicine. Part I: molecular medicine Theme Issue collection. This article describes how our experience with two university modules taught at University College London (UCL) has informed a strategy that can be applied to modules of universities across Europe and worldwide to increase the representation of women and increase diversity in the field of supercomputing.
Sensitivity analysis of a strongly-coupled human-based electromechanical cardiac model: Effect of mechanical parameters on physiologically relevant biomarkers F.Levrero-Florencio, F.Margara, E.Zacur, A.Bueno-Orovio, Z.J.Wang, A.Santiago, J.Aguado-Sierra, G.Houzeaux, V.Grau, D.Kay, M.Vázquez, R.Ruiz-Baier, B.Rodriguez,Computer Methods in Applied Mechanics and Engineering, 361 (2020) DOI: 10.1016/j.cma.2019.112762 This study presents in detail the description and implementation of a human-based coupled electromechanical modelling and simulation framework, and a high performance computing study on the sensitivity of mechanical biomarkers to key model parameters. The tools and knowledge generated enable future investigations into disease and drug action on human ventricles.
The influence of red blood cell deformability on hematocrit profiles and platelet margination B. Czaja, M. Gutierrez ,G. Závodszky, D. de Kanter, Α. Hoekstra, O. Eniola-Adefeso, PLoS Comput Biol, 16(3): e1007716, 2020, DOI: 10.1371/journal.pcbi.1007716 The deformability of red blood cells (RBCs) not only allows them to squeeze through small capillaries, yet it also impacts their flow dynamics in the plasma. We simulate varying degrees of deformability to discover that it also affects how platelets flow.
Towards Heterogeneous Multi-scale Computing on Large Scale Parallel Supercomputers S. A. Alowayyed, M. Vassaux, B. Czaja, P. V. Coveney, A. G. Hoekstra, Supercomputing Frontiers and Innovations, 2020, DOI: 10.14529/jsfi1904022 We discuss the heterogeneous multi-scale computing (HMC) pattern as a generalized method of exploiting emerging exascale computing resources, mainly concluding that, considering the subtle interplay between the macroscale model, surrogate models and micro-scale simulations, HMC provides a promising path towards exascale for many multiscale applications.
Lattice-Boltzmann interactive blood flow simulation pipeline S. S. Esfahani, X. Zhai, M. Chen, A. Amira, F. Bensaali, J. AbiNahed, S. Dakua, G. Younes, A. Baobeid, R. A. Richardson and P. V. Coveney, Int J CARS 15, pp. 629–639, 2020, DOI: 10.1007/s11548-020-02120-3 Cerebral aneurysms are one of the prevalent cerebrovascular disorders in adults worldwide and caused by a weakness in the brain artery. The most impressive treatment for a brain aneurysm is interventional radiology treatment, which is extremely dependent on the skill level of the radiologist. Hence, accurate detection and effective therapy for cerebral aneurysms still remain important clinical challenges. In this work, we have introduced a pipeline for cerebral blood flow simulation and real-time visualization incorporating all aspects from medical image acquisition to real-time visualization and steering.
Hemelb Acceleration and Visualization for Cerebral Aneurysms S. S. Esfahani, X. Zhai, M. Chen, A. Amira, F. Bensaali, J. AbiNahed, S. Dakua, G. Younes, R. A. Richardson and P. V. Coveney, 2019 IEEE International Conference on Image Processing (ICIP) Taipei, Taiwan, pp. 1376-1380, 2020, DOI: 10.1109/ICIP.2019.8803712 A weakness in the wall of a cerebral artery causing a dilation or ballooning of the blood vessel is known as a cerebral aneurysm. Optimal treatment requires fast and accurate diagnosis of the aneurysm. HemeLB is a fluid dynamics solver for complex geometries developed to provide neurosurgeons with information related to the flow of blood in and around aneurysms. On a cost efficient platform, HemeLB could be employed in hospitals to provide surgeons with the simulation results in real-time. In this work, we developed an improved version of HemeLB for GPU implementation and result visualization.
Accuracy and Precision of Alchemical Relative Free Energy Predictions With and Without Replica-Exchange S. Wan, G. Tresadem, L. Perez-Benito, H. van Vlijmen, P. V. Coveney, Advanced Theory and Simulations, 3(1), 1900195, 2020, DOI: 10.1002/adts.201900195 Advances in free energy calculations have been fostered by the integration of improved force fields, enhanced sampling methods and increased computer power. We compare the accuracy and precision of relative free energies calculated from standard TIES and two REST‐implemented approaches.

2019

Title Citation Summary
Supplementary Information from Identifying the start of a platelet aggregate by the shear rate and the cell-depleted layer B.J.M. Van Rooij; G. Závodszky; V.W. Azizi Tarksalooyeh; A.G. Hoekstra, Figshare (2019) DOI: 10.6084/m9.figshare.9831245.v1 Computer simulations were conducted to investigate red blood cell and platelet transport in high shear flows, replicating in vitro experiments in microfluidic devices prone to platelet aggregate formation. The objective was to pinpoint the initiation site of thrombus formation. Comparisons with macroscopic models revealed the necessity of cell-based modeling in microfluidic devices. Using HemoCell, a cell-based blood flow simulation framework, transport physics in these devices were explored. Simulations revealed an enlarged cell-depleted layer coinciding with platelet aggregate formation sites, where platelet concentration was higher than elsewhere. Shear rates facilitated von Willebrand factor elongation in this region. This suggests an hemodynamic environment conducive to initial platelet aggregation.
Application of the ESMACS Binding Free Energy Protocol to a Multi‐Binding Site Lactate Dehydogenase A Ligand Dataset David W. Wright, Fouad Husseini, Shunzhou Wan, Christophe Meyer, Herman van Vlijmen, Gary Tresadern, Peter V. Coveney, Advanced Theory and Simulations (2019) DOI: 10.1002/adts.201900194 Over the last two decades, fragment-based lead generation has become a widely used method in drug discovery. Computational techniques face challenges in estimating binding free energy for heterogeneous ligands. This study evaluates ensemble-based binding free energy calculation protocols (ESMACS) on ligands targeting two binding pockets in lactate dehydrogenase. Excellent statistical rankings are achieved compared to experimental results. Three methods to address entropic contributions are explored: normal mode analysis, weighted solvent accessible surface area (WSAS), and variational entropy. WSAS correlates strongly with normal mode analysis but is more computationally efficient. Variational entropy corrects exaggerated discrimination of ligands but introduces outliers.
Drug‐induced shortening of the electromechanical window is an effective biomarker for in silico prediction of clinical risk of arrhythmias Passini, E., Trovato, C., Morissette, P., Sannajust, F., Bueno-Orovio, A., & Rodriguez, B., British Journal of Pharmacology (2019) DOI: 10.1111/bph.14786 Human-based computer models offer a promising avenue for predicting drug-induced cardiac adverse events. This study investigates the potential of a cellular surrogate for the electromechanical window (EMw) as a biomarker for pro-arrhythmic cardiotoxicity. In silico drug trials were conducted for 40 reference compounds, revealing that drugs associated with Torsade de Pointes arrhythmias induce concentration-dependent EMw shortening, while safe drugs show minimal change. EMw-based predictions achieved 90% accuracy at lower concentrations compared to predictions based on repolarization abnormalities, making it a sensitive biomarker for assessing clinical pro-arrhythmic risk, especially for compounds with multichannel blocking action.
Ensemble-Based Steered Molecular Dynamics Predicts Relative Residence Time of A2A Receptor Binders Andrew Potterton, Fouad S. Husseini, Michelle W. Y. Southey, Mike J. Bodkin, Alexander Heifetz, Peter V. Coveney, Andrea Townsend-Nicholson, Journal of Chemical Theory and Computation (2019) DOI: 10.1021/acs.jctc.8b01270 We present a method utilizing atomistic ensemble-based steered molecular dynamics (SMD) to observe ligand dissociation from G protein-coupled receptors, crucial for drug discovery. This approach accurately identifies ligand-residue interactions and quantifies changes in ligand energy for both protein and water, applicable to various ligand-receptor systems. Applied to 17 ligands of the A2A adenosine receptor with known kinetic binding data, our method demonstrates a strong correlation (R2 = 0.79) between computationally calculated changes in water-ligand interaction energy and experimentally determined residence time. This semi-empirical approach offers rapid and precise determination of drug-target relative residence time, facilitating optimization in drug discovery programs.
Improving the clinical understanding of hypertrophic cardiomyopathy by combining patient data, machine learning and computer simulations: A case study Lyon, A., Mincholé, A., Bueno-Orovio, A., & Rodriguez, B., Morphologie : bulletin de l’Association des anatomistes (2019) DOI: 10.1016/j.morpho.2019.09.001 This paper illustrates the effective application of computational techniques in clinically-relevant contexts, particularly in hypertrophic cardiomyopathy (HCM), a prevalent genetic cardiac disease. Through a combination of electrocardiogram and imaging data, machine learning, and high-performance computing simulations, the study identified four distinct phenotypes in HCM, each associated with varying levels of arrhythmic risk. Furthermore, the research proposed two potential mechanisms contributing to the heterogeneity of HCM manifestation. These findings facilitate improved patient stratification and enhance understanding of disease mechanisms, thereby advancing personalized management and treatment approaches for HCM patients.
The effect of boundary and loading conditions on patient classification using finite element predicted risk of fracture Zainab Altai, Muhammad Qasim, Xinshan Li, Marco Viceconti, Clinical Biomechanics (2019) DOI: 10.1016/j.clinbiomech.2019.06.004 Osteoporotic proximal femoral fractures due to falls pose significant health challenges in aging populations. Finite element models estimating bone strength offer potential for fracture risk prediction, but accuracy varies. This study explores different side fall boundary conditions in predicting fracture risk among postmenopausal women. Results show that the Contact model outperforms others, yielding the highest stratification power (0.82), followed closely by Multi-point constraints and Linear models (0.80 each). Notably, both Contact and Multi-point constraints models predict high strains in diverse proximal femur locations, including the greater trochanter, an underreported area.
Characterising GPCR–ligand interactions using a fragment molecular orbital-based approach A. Heifetz, T. James, M. Southey, I. Morao, M. Aldeghi, L. Sarrat, D. G. Fedorov, M. J. Bodkin, A. Townsend-Nicholson, Current Opinion in Structural Biology 55, 85-92 (2019) DOI: 10.1016/j.sbi.2019.03.021 There has been fantastic progress in solving GPCR crystal structures. However, the ability of X-ray crystallography to guide the drug discovery process for GPCR targets is limited by the availability of accurate tools to explore receptor–ligand interactions. Visual inspection and molecular mechanics approaches cannot explain the full complexity of molecular interactions. Quantum mechanical approaches (QM) are often too computationally expensive, but the fragment molecular orbital (FMO) method offers an excellent solution that combines accuracy, speed and the ability to reveal key interactions that would otherwise be hard to detect.
Location-Specific Comparison Between a 3D In-Stent Restenosis Model and Micro-CT and Histology Data from Porcine In Vivo Experiments P. S. Zun, A. J. Narracott, C. Chiastra, J. Gunn and A. G. Hoekstra, Cardiovascular Engineering and Technology 10, 568–582, (2019) DOI: 10.1007/s13239-019-00431-4 Coronary artery restenosis is an important side effect of percutaneous coronary intervention. Computational models can be used to better understand this process. We report on an approach for validation of an in silico 3D model of in-stent restenosis in porcine coronary arteries and illustrate this approach by comparing the modelling results to in vivo data for 14 and 28 days post-stenting. The approach presented here provides a very detailed, location-specific, validation methodology for in silico restenosis models. The model was able to closely match both histology datasets with a single set of parameters. Good agreement was obtained for both the overall amount of neointima produced and the local distribution.
Identifying the start of a platelet aggregate by the shear rate and the cell-depleted layer B. J. M. van Rooij, G. Závodszky, V. W. Azizi Tarksalooyeh and A. G. Hoekstra, J. R. Soc. Interface 26, 20190148 (2019) DOI: 10.1098/rsif.2019.0148 Computer simulations were performed to study the transport of red blood cells and platelets in high shear flows, mimicking earlier published in vitro experiments in microfluidic devices with high affinity for platelet aggregate formation. The goal is to understand and predict where thrombus formation starts. We hypothesize that the enlarged cell-depleted layer combined with a sufficiently large platelet flux and sufficiently high shear rates result in an haemodynamic environment that is a preferred location for initial platelet aggregation.
High arrhythmic risk in antero-septal acute myocardial ischemia is explained by increased transmural reentry occurrence H. Martinez-Navarro, A. Mincholé, A. Bueno-Orovio, B. Rodriguez, Scientific Reports, 9, 16803 (2019), DOI: 10.1038/s41598-019-53221-2 Acute myocardial ischemia is a precursor of sudden arrhythmic death. Variability in its manifestation hampers understanding of arrhythmia mechanisms and challenges risk stratification. Our aim is to unravel the mechanisms underlying how size, transmural extent and location of ischemia determine arrhythmia vulnerability and ECG alterations. High performance computing simulations using a human torso/biventricular biophysically-detailed model were conducted to quantify the impact of varying ischemic region properties, including location (LAD/LCX occlusion), transmural/subendocardial ischemia, size, and normal/slow myocardial propagation. The technology and results presented can inform safety and efficacy evaluation of anti-arrhythmic therapy in acute myocardial ischemia.
Continuum model for flow diverting stents in 3D patient-specific simulation of intracranial aneurysms S. Li, B. Chopard and J. Latt, Journal of Computational Science, 38, 101045 (2019), DOI: 10.1016/j.jocs.2019.101045 The present work extends the framework of screen-based flow diverter model (SFDM) to 3D flows and validates it using actual medical flow diverters in patient specific aneurysms. The numerical tests show that the SFDM can reproduce the results of direct numerical simulation both qualitatively and quantitatively with high precision, and are capable of reducing the simulation time by an order of magnitude or more. The article discusses the procedure required to deploy the model for a given stent and artery.
DeltaDelta neural networks for lead optimization of small molecule potency J. Jiménez-Luna, L. Pérez-Benito, G. Martínez-Rosell, S. Sciabola, R. Torella, G. Tresadern and G. De Fabritiis, Chemical Science, 47, (2019), DOI: 10.1039/C9SC04606B The capability to rank different potential drug molecules against a protein target for potency has always been a fundamental challenge in computational chemistry due to its importance in drug design. While several simulation-based methodologies exist, they are hard to use prospectively and thus predicting potency in lead optimization campaigns remains an open challenge. Here we present the first machine learning approach specifically tailored for ranking congeneric series based on deep 3D-convolutional neural networks.
Bridging the computational gap between mesoscopic and continuum modeling of red blood cells for fully resolved blood flow C. Kotsalos, J. Latt and B. Chopard J. Comput. Phys, 2019, DOI: 10.1016/j.jcp.2019.108905 A computational framework for the simulation of blood flow with fully resolved red blood cells using a modular approach that consists of a lattice Boltzmann solver, a novel finite element based solver and an immersed boundary method.
Big data: the end of the scientific method? S. Succi and P. V. Coveney, Philos. T. R. Soc. A., 2019, DOI: 10.1098/rsta.2018.0145 We argue that the boldest claims of big data (BD) are in need of revision and toning-down, in view of a few basic lessons learned from the science of complex systems. We propose a synergistic merging, as opposed to antagonism, between BD and mechanistic theory, for a new scientific paradigm capable of overcoming some of the major barriers confronted by the modern scientific method originating with Galileo.
Red blood cell and platelet diffusivity and margination in the presence of cross-stream gradients in blood flows G. Závodszky, B. van Rooij, B. Czaja, V. Azizi, D. de Kanter and A. G. Hoekstra, Physics of Fluids, 2019, DOI: 10.1063/1.5085881 The radial distribution of cells in blood flow inside vessels is highly non-homogeneous, leading to complex fluid dynamics of which the mechanisms are not fully understood. Single-cell, cell-pair, and large-scale many-cell simulations are performed using a validated numerical model in order to gain insight into this complexity.
Multiple Aneurysms AnaTomy CHallenge 2018 (MATCH)—phase II: rupture risk assessment P. Berg, S. Voß, G. Janiga, et al. Int. J. CARS., 2019, 10.1007/s11548-019-01986-2 Assessing the rupture probability of intracranial aneurysms remains challenging. Therefore, hemodynamic simulations are increasingly applied toward supporting physicians during treatment planning. However, due to several assumptions, the clinical acceptance of these methods remains limited.
β-Adrenergic Receptor Stimulation and Alternans in the Border Zone of a Healed Infarct: An ex vivo Study and Computational Investigation of Arrhythmogenesis J. Tomek, G. Hao, M. Tomková, A. Lewis, C. Carr, D. J. Paterson, B. Rodriguez, G. Bub and N. Herring, Front. Physiol., 2019, 10.3389/fphys.2019.00350 Following myocardial infarction (MI), the myocardium is prone to calcium-driven alternans, which typically precedes ventricular tachycardia and fibrillation. We hypothesize that the infarct border zone is most vulnerable to alternans, that β-adrenergic receptor stimulation can suppress this, and investigate the consequences in terms of arrhythmogenic mechanisms.
Predicting Activity Cliffs with Free-Energy Perturbation L. Pérez-Benito, N. Casajuana-Martin, M. Jiménez-Rosés, H. van Vlijmen and G. Tresadern, J. Chem. Theory Comput., 2019, 10.1021/acs.jctc.8b01290 Activity cliffs (ACs) are an important type of structure–activity relationship in medicinal chemistry where small structural changes result in unexpectedly large differences in biological activity. Being able to predict these changes would have a profound impact on lead optimization of drug candidates.
Mechanisms Underlying Allosteric Molecular Switches of Metabotropic Glutamate Receptor 5 C. L. del Torrent, N. Casajuana-Martin, L. Pardo, G. Tresadern and L. Pérez-Benito, J. Chem. Inf. Model., 2019, 10.1021/acs.jcim.8b00924 The metabotropic glutamate 5 (mGlu5) receptor is a class is implicated in several central nervous system disorders, making it a popular drug discovery target, however the origins of its effect are not understood, causing difficulties in a drug discovery context. We juxtapose experimental and simulation results to investigate these effects./td>
Improved biomechanical metrics of cerebral vasospasm identified via sensitivity analysis of a 1D cerebral circulation model A. Melis, F. Moura, I. Larrabide, K. Janot, R. H. Clayton, A. P. Narata and A. Marzo J. Biomech., 201910.1016/j.jbiomech.2019.04.019 Cerebral vasospasm (CVS) is a life-threatening condition that occurs in a large proportion of those affected by subarachnoid haemorrhage and stroke. The aim of this study is to identify alternative biomarkers that could be used to diagnose CVS.
Computational Drug Design Applied to the Study of Metabotropic Glutamate Receptors C. L. del Torrent, L. Pérez-Benito and G. Tresadern, Molecules, 201910.3390/molecules24061098 Metabotropic glutamate receptors are a family of eight GPCRs that are attractive drug discovery targets to modulate glutamate action and response. We review the application of computational methods to the study of this family of receptors.
The effect of boundary and loading conditions on patient classification using finite element predicted risk of fracture Z. Altai, M. Qasim, X. Li, and M. Viceconti, Clin. Biomech., 2019, DOI: 10.1016/j.clinbiomech.2019.06.004 Osteoporotic proximal femoral fractures associated to falls are a major health burden in the ageing society. This study investigates the ability of the Finite Element predicted strength in classifying fracture and non-fractured cases.
Investigating the complex arrhythmic phenotype caused by the gain-of-function mutation KCNQ1-G229D X. Zhou, A. Bueno-Orovio, R. J. Schilling, C. Kirby, C. Denning, D. Rajamohan, K. Burrage, A. Tinker, B. Rodriguez and S. HarmerFront. Physiol., 2019, DOI: 10.3389/fphys.2019.00259 A cardiac electrophysiological disorder that can result in sudden cardiac death is caused by the mutation of protein KCNQ1. We investigate the ionic, cellular and tissue mechanisms underlying the complex phenotype of KCNQ1 mutation using computer modeling and simulations informed by in vitro measurements.
Application of the ESMACS binding free energy protocol to a multi-binding site lactate dehydogenase A ligand dataset D. W. Wright, F. Husseini, S. Wan, C. Meyer, H. van Vlijmen, G. Tresadern, P. V. Coveney, Advanced Theory and Simulations, 2019, DOI: 10.1002/adts.201900194 Fragment-based lead generation has become a common, mature approach to identify tractable starting points in chemical space for the drug discovery process. This approach naturally involves the study of the binding properties of highly heterogeneous ligands. We evaluate the performance of our range of ensemble simulation based binding free energy calculation protocols, called ESMACS, by comparison to experimental results.
A New Pathology in the Simulation of Chaotic Dynamical Systems on Digital Computers A. Potterton, F. Husseini, M. Southey, M. Bodkin, A. Heifetz, P. V. Coveney, A. Townsend-Nicholson, Advanced Theory and Simulations, 1900125, 2019, DOI:10.1002/adts.201900125 Systematic distortions are uncovered in the statistical properties of chaotic dynamical systems when represented and simulated on digital computers using standard IEEE floating-point numbers. The analysis indicates that the pathology described, which cannot be mitigated by increasing the precision of the floating point numbers, is a presentative example of a deeper problem in the computation of expectation values for chaotic systems.
Ensemble-Based Steered Molecular Dynamics Predicts Relative Residence Time of A2A Receptor Binders A. Potterton, F. Husseini, M. Southey, M. Bodkin, A. Heifetz, P. V. Coveney, A. Townsend-Nicholson, Journal of Chemical Theory and Computation, 15 (5), 3316–3330 2019, 10.1021/acs.jctc.8b01270 We present a novel computational method for the reliable prediction of relative drug-target residence time. From ensemble-based steered molecular dynamics simulations, the change in energy between the ligand and water during dissociation is obtained. This energy correlates strongly to the associated experimental residence times of receptor ligands.
Application of ESMACS binding free energy protocols to diverse datasets:Bromodomain-containing protein 4 D. W. Wright, S. Wan, C. Meyer, H. van Vlijmen, G. Tresadern, P. V. Coveney, Scientific Reports, 9, 6017, 2019, 10.26434/chemrxiv.7327019 We investigate the robustness of our ensemble molecular dynamics binding free energy protocols, known as ESMACS, to different choices of forcefield, starting structure and analysis. We examine the influence of multiple trajectories, explicit water molecules and estimates of the entropic contribution to the binding free energy.
PathwayMap: Molecular Pathway Association with Self-Normalizing Neural Networks J. Jiménez, D. Sabbadin, A. Cuzzolin, G. Martínez-Rosell, J. Gora, J. Manchester, J. Duca, G. De Fabritiis, J. Chem. Inf. Model., 2019, 10.1021/acs.jcim.8b00711 Drug discovery suffers from high attrition because compounds initially deemed as promising can later show ineffectiveness or toxicity resulting from a poor understanding of their activity profile. Here, we describe a deep self-normalizing neural network model for the prediction of molecular pathway association and evaluate its performance.
Shape-Based Generative Modeling for de Novo Drug Design M. Skalic, J. Jiménez, D. Sabbadin, G. De Fabritiis, J. Chem. Inf. Model., 2019, 10.1021/acs.jcim.8b00706 A machine learning approach to generate novel molecules starting from a seed compound, its three-dimensional (3D) shape, and its pharmacophoric features. The pipeline draws inspiration from generative models used in image analysis and represents a first example of the de novo design of lead-like molecules guided by shape-based features.
Machine Learning of Coarse-Grained Molecular Dynamics Force Fields J. Wang, S. O, C. Wehmeyer, A. Pérez, N. E. Charron, G. de Fabritiis, F Noé, and C. Clementi ACS Cent. Sci., 2019, 10.1021/acscentsci.8b00913 Coarse-grained models offer the advantage of spanning greater length- and timescales in simulations at the cost of precision. In this paper, we reformulate coarse-graining as a supervised machine learning problem and use statistical learning theory to compare between different models.
Advanced HPC-based Computational Modeling in Biomechanics and Systems Biology M. Vázquez, P. V. Coveney, H. E. Grecco, A. Hoekstra, B. Chopard (editors), Frontiers in Physiology, Frontiers in Applied Mathematics and Statistics and Frontiers in Bioengineering and Biotechnology, e-Book 2019 A collection of articles on how advanced high-performance computing (HPC) methods can facilitate models of biomechanics and systems biology.
Characterising GPCR-ligand interactions using a fragment molecular orbital-based approach. A. Heifetz, T. James, M. Southey, I. Morao, M. Aldeghi, L. Sarrat, D. G. Fedorov, M. J. Bodkin, A. Townsend-Nicholson Curr. Opin. Struct. Biol., 2019, 10.1016/j.sbi.2019.03.021 There has been fantastic progress in solving GPCR crystal structures. However, the ability of X-ray crystallography to guide the drug discovery process for GPCR targets is limited. Integration of GPCR crystallography or homology modelling with the fragment molecular orbital (FMO) method reveals unprecedented atomistic details.
Computational prediction of GPCR oligomerization. A. Townsend-Nicholson, N. Altwaijry, A. Potterton, I. Morao, A. Heifetz.Curr. Opin. Struct. Biol., 2019, 10.1016/j.sbi.2019.04.005 Ensemble-based computational methods based on structurally determined dimers, coupled with a computational workflow that uses quantum mechanical methods to analyze the chemical nature of the molecular interactions at a GPCR dimer interface, will generate the reproducible and accurate predictions needed to predict previously unidentified GPCR dimers and to inform future advances in our ability to understand and begin to precisely manipulate GPCR oligomers in biological systems.
Toward Full GPU Implementation of Fluid-Structure Interaction J. Bény, C. Kotsalos, J. LattIEEE, 2019, 10.1109/ISPDC.2019.000-2 Fluid-structure interaction (FSI) is a notoriously difficult topic in the fields of computational fluid dynamics (CFD) and finite element analysis (FEA), as it requires the deployment of a coupling framework between two different numerical methodologies. In this article, we present a strategy for deploying a FSI system in the many-thread framework of general purpose Graphics Processing Units (gpGPUs).

2018

Title Citation Summary
Investigating the mechanical response of paediatric bone under bending and torsion using finite element analysis Zainab Altai, Marco Viceconti, Amaka C. Offiah, Xinshan Li, Biomechanics and Modeling in Mechanobiology (2018) DOI: 10.1007/s10237-018-1008-9 Fractures are common in pediatric injuries, comprising 25% of cases, and can be accidental or inflicted. Distinguishing between them is often subjective for clinicians due to limited evidence-based data on pediatric bone strength. This study aims to investigate the response of pediatric femora to bending and torsional loads using CT-based finite element analysis, and establish a relationship between bone strength and age/body mass. Thirty post-mortem CT scans of children aged 0-3 years were analyzed. Results indicate that bone strength increases with age and body mass, with failure moments ranging from 0.8-27.9 Nm for bending and 1.0-31.4 Nm for torsion. This study provides novel insights into infant bone strength, aiding the development of child safety systems and more accurate computer models for pediatric care.
Are CT-Based Finite Element Model Predictions of Femoral Bone Strengthening Clinically Useful? Marco Viceconti, Muhammad Qasim, Pinaki Bhattacharya, Xinshan Li, Current Osteoporosis Reports (2018) DOI: 10.1007/s11914-018-0438-8 Reviewing existing literature, this study compares the accuracy of areal bone mineral density (DXA-aBMD) derived from dual X-ray absorptiometry with subject-specific finite element models (QCT-SSFE) derived from quantitative computed tomography in predicting bone strength, both experimentally and clinically. It evaluates their efficacy in discriminating and predicting bone strength, alongside basic cost-effectiveness analyses. The research explores potential advantages of employing QCT-SSFE over DXA-aBMD in clinical studies assessing femoral strength and predicting hip fracture risk in patients with low bone mass.
Understanding Malaria Induced Red Blood Cell Deformation Using Data-Driven Lattice Boltzmann Simulations J. S. Y. Tan, G. Závodszky and P. M. A. Sloot, Computational Science – ICCS 2018, 2018, DOI: 10.1007/978-3-319-93698-7_30 Malaria remains a deadly disease that affected millions of people in 2016. Among the five Plasmodium (P.) parasites which contribute to malaria in humans, P. falciparum is lethal and responsible for the majority of cases. In this study we use lattice Boltzmann simulations to investigate infected red blood cells.
Strategies of data layout and cache writing for input-output optimization in high performance scientific computing: Applications to the forward electrocardiographic problem L. Cardone-Noott, B. Rodriguez, A. Bueno-Orovio PloS One, 2018, DOI: 10.1371/journal.pone.0202410 Input-output (I/O) optimization at the low-level design of data layout on disk drastically impacts the efficiency of high performance computing (HPC) applications. We present a novel low-level data layout for HPC applications, fully independent of the number of dimensions in the dataset.
Are CT-Based Finite Element Model Predictions of Femoral Bone Strengthening Clinically Useful? M. Viceconti, M. Qasim, P. Bhattacharya and X.Li, Curr. Osteoporosis Rep., 2018, DOI: 10.1007/s11914-018-0438-8 This study reviews the available literature to compare the accuracy of various imaging techniques and combinations of such with computer models in predicting bone strength.
Ensemble-based replica exchange alchemical free energy methods: the effect of protein mutations on inhibitor binding A. P. Bhati, S. Wan, and P. V. Coveney, J. Chem. Theory Comput., 2018, 10.1021/acs.jctc.8b01118 The accurate prediction of the binding affinity changes of drugs caused by protein mutations is a major goal in clinical personalized medicine. We have developed TIES, an ensemble-based free energy approach which yields accurate, precise, and reproducible binding affinities.
Concurrent and Adaptive Extreme Scale Binding Free Energy Calculations J. Dakka, K. Farkas-Pall, M. Turilli, D. W. Wright, P. V. Coveney, S. Jha, ArXiv, 2018, arxiv.org/abs/1801.01174 The efficacy of drug treatments depends on how tightly the drug’s small molecules bind to their target proteins. We introduce the high-throughput binding affinity calculator (HTBAC), a molecular dynamics framework, as a step towards rapid and accurate quantification of drug-protein interactions and towards surmounting the grand challenge of computational chemistry which could revolutionize drug design and provide the platform for patient-specific medicine.
Identifying inter-helical interactions involved in GPCR structure-function and the forces that determine ligand residence time. A. Heifetz, A. Potterton, I. Morao, T. James, M. Southey, D. Fedorov, M.Bodkin, A. Townsend-NicholsonAbstracts of Papers of the American Chemical Society, 2018, 10.1016/j.sbi.2019.04.005 There has been a recent and prolific expansion in the number of GPCR crystal structures being solved: in both active and inactive forms and in complex with ligand, with G protein and with each other. Ensemble-based computational methods based on structurally determined dimers, coupled with a computational workflow that uses quantum mechanical methods to analyze the chemical nature of the molecular interactions at a GPCR dimer interface, will generate the reproducible and accurate predictions needed to predict previously unidentified GPCR dimers and to inform future advances in our ability to understand and begin to precisely manipulate GPCR oligomers in biological systems.
Fully coupled Fluid-electro-mechanical model of the human heart for supercomputers Santiago A, Zavala‐Aké M, Aguado‐Sierra J, Doste R, Gómez S, Arís R, Cajas J C, Casoni E, Vázquez M, Int. J. Numer. Meth. Biomed. Engng., 2018, 10.1002/cnm.3140 The first fluid-electro-mechanical model of the human heart is presented. Such a model allows to analyse all the physics involved in the heartbeat in a whole-heart geometrical description. Such an integrative description of the heartbeat provides to a better understanding of complex cardiopathies.
Implications of bipolar voltage mapping and magnetic resonance imaging resolution in biventricular scar characterisation after myocardial infarction M. López-Yunta, D. G. León, J. M. Alfonso-Almazán, M. Marina-Breysse, J. G. Quintanilla, J. Sánchez-González, C. Galán-Arriola, V. Cañadas-Godoy, D. Enríquez-Vázquez, C. Torres, B. Ibáñez, J. Pérez-Villacastín, N. Pérez-Castellano, J. Julie, M. Vázquez, J. Aguado-Sierra, D. Filgueiras-Rama, Europace, 2018, 10.1093/europace/euy192 Scar characterization using different cardiac imaging modalities is a common clinical approach to stratify the risk of ventricular arrhythmia and identify potential target areas for catheter-based ablation after myocardial infraction. On a sample of pigs we perform a study of the differences in scar characterization using bipolar voltage mapping compared with state-of-the-art in vivo delayed gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) imaging and ex vivo T1 mapping.
Influence of fiber connectivity in simulations of cardiac biomechanics D. Gil, R. Aris, A. Borras, E. Ramirez, R. Sebastian, M. Vazquez, International Journal of Computer Assisted Radiology and Surgery, 2018, 10.1007/s11548-018-1849-9 We explore the influence of the fiber distribution that is used in cardiac simulations, and show that fibers extracted from experimental data produce functional scores closer to healthy ranges than mathematical models. We conclude that deep knowledge of the cardiac fiber field is important to achieve more realistic results in computational modelling.
PlayMolecule BindScope: large scale CNN-based virtual screening on the web M. Skalic, G. Martínez-Rosell, J. Jiménez, G. De Fabritiis, Bioinformatics, 2018, 10.1093/bioinformatics/bty758 A deep learning method to distinguish active from non-active ligands for large-scale virtual screening of drug candidates, which is made available in an easy-to-use web platform. Bindscope accelerates the initial stages of drug-discovery, speeding up the time required for novel drug development.
Numerical Investigation of the Effects of Red Blood Cell Cytoplasmic Viscosity Contrasts on Single Cell and Bulk Transport Behaviour M. de Haan, G. Závodszky, V. Azizi, A. G. Hoekstra, Applied Sciences, 8(9), 1616, 2018, 10.3390/app8091616 Assessing the influence of internal viscosity of red blood cells on the properties of whole blood.
Cell-resolved blood flow simulations of saccular aneurysms: effect of pulsatility and aspect ratio B. Czaja, G. Závodszky, V. Azizi Tarksalooyey, A. G. Hoekstra, Journal of the Royal Society Interface, 2018, 10.1098/rsif.2018.0485 How geometrical properties of aneurysms influence the transport of platelets into these aneurysms.
Inflow and outflow boundary conditions for 2D suspension simulations with the immersed boundary lattice Boltzaman method V. Azizi Tarksalooyeh, G. Závodszky, B. J. M. van Rooij, A. G. Hoekstra, Computers & Fluids, 2018, 10.1016/j.compfluid.2018.04.025 An algorithm for establishing inlet and outlet conditions for simulations of blood flow, explicitly following every single red blood cell in the blood.
Patterns for High Performance Multiscale Computing S. Alowayyed, T. Piontek, J. L. Suter, O. Hoenen, D. Groen, O. Luk, B. Bosak, P. Kopta, K. Kurowski, O. Perks, K. Brabazon, V. Jancauskas, D. Coster, P.V. Coveney, A.G. Hoekstra, Future Generation Computer Systems, 91, 335-346 2018, 10.1016/j.future.2018.08.045 A generic implementation to effectively use supercomputers for multiscale models, including biomedical ones, focusing on automated execution on High-Performance Computers (HPC), delivering performance benefits from both end-user and HPC-system-level perspectives.
Uncertainty Quantification of a Multiscale Model for In-Stent Restenosis A. Nikishova, L. Veen, P. Zun, A. G. Hoekstra, Cardiovascular Engineering and Technology, 1-4, 2018, 10.1007/s13239-018-00372-4 Estimating the uncertainties in models of scar tissue prediction in treatment of stenosed coronary arteries.
Parameter Estimation of Platelets Deposition: Approximate Bayesian Computation with High Performance Computing R. Dutta, B. Chopard, J. Latt, F. Dubois, K.Z. Boudjeltia, A. Mira, Front Physiol, 2018, 10.3389/fphys.2018.01128 Existing clinical tests to detect cardio/cerebrovascular diseases (CVD) are ineffectual, in part because they do not consider different stages of platelet activation or the dynamics involved in platelet interactions. Here, we devise a Bayesian inferential scheme for the estimation of these dynamics and support that our approach can be used to build a new generation of personalized platelet functionality tests for CVD detection.
LigVoxel: Inpainting binding pockets using 3D-convolutional neural networks M. Skalic, A. Varela-Rial, J. Jimenez, G. Martinez-Rosell, G. De Fabritiis, Bioinformatics , 2018, 10.1093/bioinformatics/bty583 A proposition of a purely data driven, structure-based approach for imaging ligands as spatial fields in target protein pockets. This computational method speeds up the design of novel molecules, and assists the improvement of potential drug candidates.
Simulations meet machine learning in structural biology A. Perez, G. Martinez-Rosell, G. De Fabritiis, Current Opinion in Structural Biology , 2018, 10.1016/j.sbi.2018.02.004 Review article on the impact of the latest machine learning methods in molecular dynamics simulations and structural biology and a future perspective on the field.
Left Ventricular Trabeculations Decrease the Wall Shear Stress and Increase the Intra-Ventricular Pressure Drop in CFD Simulations F. Sacco, B. Pain, O. Lehmkuhl, T.L. Iles, P.A. Iaizzo, G. Houzeaux, M. Vazquez, C. Butakoff, J. Aguado-Sierra, Frontiers in Physiology, 2018, 10.3389/fphys.2018.00458 An examination of the impact of the trabeculae and papillary muscles of the heart on blood flow using high performance computing (HPC) and models reconstructed from high-resolution magnetic resonance images of ex-vivo human hearts.
Evaluating the roles of detailed endocardial structures on right ventricular haemodynamics by means of CFD simulations F. Sacco, B. Pain, O. Lehmkuhl, T.L. Iles, P.A. Iaizzo, G. Houzeaux, M. Vazquez, C. Butakoff, J. Aguado-Sierra, International Journal for Numerical Methods in Biomedical Engineering, 2018, 10.1002/cnm.3115 A detailed computational fluid dynamics study is performed in four human male and female heart geometries, with wall shear stress, pressure drop and turbulence as calculated quantities. We find that neglecting internal structures in cardiac models may lead to inaccurate conclusions about the aforementioned quantities.
A mechanistic model for predicting cell surface presentation of competing peptides by MHC class I molecules D. S. M. Boulanger, R. C. Eccleston, A. Phillips, P. V. Coveney, T. Elliott, N. Dalchau, Frontiers Immunology, 2018, 10.3389/fimmu.2018.01538 We develop and experimentally verify a mechanistic model for presentation of peptides to major histocompatibility complex-I (MHC-I) molecules. The resulting model can be used to predict key steps in the processing of intracellular peptides, which play an important role in inflammatory immune responses to viruses and cancer.
Validation of patient-specific cerebral blood flow simulation using transcranial Doppler measurements D. Groen, R. A. Richardson, R. Coy, U. D. Schiller, H. Chandrashekar, F. Robertson, P. V. Coveney, Frontiers Physiology, 2018, 10.3389/fphys.2018.00721 We show that the HemeLB simulation code is able to reproduce velocities measured using transcranial Doppler in a cerebral artery. HemeLB allows the study of blood flow systems such as aneurysms, and by improving our understanding through simulation we can improve and discover new treatments and preventions.
PolNet: A Tool to Quantify Network-Level Cell Polarity and Blood Flow in Vascular Remodeling M. O. Bernabeu, M. L. Jones, R. W. Nash, A. Pezzarossa, P. V. Coveney, H. Gerhardt, and C. A. Franco, Biophysical Journal, 114 (9), 2052-2058 2018, 10.1016/j.bpj.2018.03.032 PolNet is an open-source software tool for the study of blood flow and cell-level biological activity during vessel morphogenesis. PolNet will be a powerful analysis method to address the complexity of endothelial cell biology at the network level in intact organs.
Uncertainty Quantification in Alchemical Free Energy Methods A. Bhati, S. Wan, Y. Hu, B. Sherborne, P. V. Coveney, Journal of Chemical Theory and Computation, 14 (6), 2867-2880 2018, 10.1021/acs.jctc.7b01143 This study provides a systematic approach to uncertainty quantification based on ensemble simulations, which is generally applicable to all free energy calculation methods that draw on classical molecular dynamics.
Load balancing of parallel cell-based blood flow simulations S. Alowayyed, G. Závodszky, V. Azizi and A. G. Hoekstra J. Comput. Sci., 2018, 10.1016/j.jocs.2017.11.008 The non-homogeneous distribution of computational costs is often challenging to handle in highly parallel applications. Using a methodology based on fractional overheads, we formulate and validate a model for the fractional load imbalance and compare it with other sources of overhead, in particular the communication overhead.
Investigating the mechanical response of paediatric bone under bending and torsion using finite element analysis Z. Altai, M. Viceconti, A. C. Offiah, X. Li, Biomechanics and Modeling in Mechanobiology, 2018, 10.1007/s10237-018-1008-9 This is the first study to quantitatively analyse the infant bone strength under different loading conditions. The results are precious to the research into childhood bone diseases and fractures, especially those in the very young age range. This research could contribute to new medical treatments and preventions of bone diseases and fractures.
Modeling Patient-Specific Magnetic Drug Targeting Within the Intracranial Vasculature A. Patronis, R. A. Richardson, S. Schmieschek, B. J. N. Wylie, R. W. Nash, P. V. Coveney, Frontiers in Physiology, 2018, 10.3389/fphys.2018.00331 We have applied our magnetic-drug targeting model to paramagnetic-nanoparticle-laden flows in a geometry obtained from an MRI scan and we note a strong dependence of the particle density on the strength of the magnetic forcing and the velocity of the background fluid flow.
Predicting Binding Free Energies of PDE2 Inhibitors. The Difficulties of Protein Conformation L. Pérez-Benito, H. Keränen, H. van Vlijmen & G. Tresadern, Scientific Reports, 8, 4833, 2018, 10.1038/s41598-018-23039-5 his work shows that computational methods can help to predict the interaction energy between candidate drug molecules and their target protein, although the inherent flexibility of proteins makes this more difficult than expected. Large computational resources are required for these accurate calculations, but they could greatly enhance the role of computational assisted drug design.
KDEEP: Protein-Ligand Absolute Binding Affinity Prediction via 3D-Convolutional Neural Networks J. Jimenez, M. Skalic, G. Martinez-Rosell, G. De Fabritiis, J. Chem. Inf. Model.,, 52 (8), 287-296, 2018, 10.1021/acs.jcim.7b00650 A deep learning method to predict how strongly ligands bind to proteins, providing fast predictions with similar accuracy compared with other state-of-the-art methods.
Computational Methods for GPCR Drug Discovery A. Heifetz, Springer, 1705, 2018, 10.1007/978-1-4939-7465-8 This book provides a unique overview of modern computational strategies and techniques employed in the field of G protein-coupled receptors (GPCRs) drug discovery, including structure- and ligand-based approaches and cheminformatics.
Synergistic Use of GPCR Modeling and SDM Experiments to Understand Ligand Binding A. Potterton, A. Heifetz, A. Townsend-Nicholson, Methods Mol Biol., 1705, 335-343, 2018, 10.1007/978-1-4939-7465-8_15 We describe a protocol by which historic ligand binding data and computational models that the former inspire may be used together to understand ligand binding.
Unlocking data sets by calibrating populations of model to data density: A study in atrial electrophysiology B. A. J. Lawson, C. C. Drovandi, N. Cusimano, P. Burrage, B. Rodriguez, K. Burrage, Science Advances, 4 (1), 2018, 10.1126/sciadv.1701676 This study describes the ability of a parameter sampling algorithm to produce populations of models calibrated to data distributions.
Computational Methods Used in Hit-to-Lead and Lead Optimization Stages of Structure-Based Drug Discovery A. Heifetz, M. Southey, I. Morao, A. Townsend-Nicholson, M. Bodkin, Methods Mol Biol., 1705, 375-394 2018, 10.1007/978-1-4939-7465-8_19 This book chapter explores drug interaction with proteins and the modern computational strategy of Hit-to-Lead and Lead Optimization Stages of Structure-Based G protein-coupled receptors (GPCR) Drug Discovery. By exploring and developing such computational methods, we move closer to a future where drug design is regularly informed by simulations.
The application of the screen model for stents in cerebral aneurysms S. Li, J. Latt, B. Chopard, Computer Fluids., 172, 651-660 2018, 10.1016/j.compfluid.2018.02.007 Cerebral aneurysms can be treated by inserting a flow-diverter in the parent artery. This study advances a previous model incorporating a flow-diverter by validating a proposed model for complex flow problems and quantifying the benefit of the approach in term of computing speed and accuracy.

2017

Title Citation Summary
Dynamic and Kinetic Elements of µ-Opioid Receptor Functional Selectivity A. Kapoor, G. Martinez-Rosell, D. Provasi, G. Fabritiis, M Filizola, Scientific Reports, 7, 11255 2017, 10.1038/s41598-017-11483-8 An application of high-throughput methods and kinetic analysis to an important protein family for drug-targeting (GPCR), related to mental-health diseases and cancer. These results provide important insights to develop better drugs for GPCR related diseases.
Modelling variability in cardiac electrophysiology: a moment-matching approach E. Tixier, D. Lombardi, B. Rodriguez, J-F. Gerbeau J. R. Soc. Interface., 2017, 10.1098/rsif.2017.0238 We present a method which serves the general purpose of estimating cardiac model parameters from a set of measurements of electrophysiology. The proposed approach may be a new way to investigate features observed in electrophysiology that are experimentally difficult to assess and may have potentially important implications in drug safety pharmacology.
Human In Silico Drug Trials Demonstrate Higher Accuracy than Animal Models in Predicting Clinical Pro-Arrhythmic Cardiotoxicity E. Passini, O. J. Britton, H. R. Lu, J. Rohrbacher, A. N. Hermans, D. J. Gallacher, R. J. H. Greig, A. Bueno-Orovio, B. Rodriguez Front. Physiol., 2017, 10.3389/fphys.2017.00668 Early prediction of damage to the heart muscle is critical for drug development. Our study demonstrates that human in silico drug trials constitute a powerful methodology for the prediction of early muscle damage, with better inference potential than equivalent animal models.
Model for pressure drop and flow deflection in the numerical simulation of stents in aneurysms. International journal for numerical methods in biomedical engineering S. Li, J. Latt, B. Chopard, International Journal for Numerical Methods in Biomedical Engineering, 2017, 10.1002/cnm.2949 Cerebral aneurysms can be treated by inserting a flow-diverter in the parent artery. This study proposes simulations which incorporate a description of a flow-diverter, including patient specific cases. Through simulations we advance our understanding of this aneurysm treatment, leading to improvements and new approaches.
δ‐cells and β‐cells are electrically coupled and regulate α‐cell activity via somatostatin L. J. B. Briant, T. M. Reinbothe., I. Spiliotis, C. Miranda, B. Rodriguez and P. Rorsman J. Physiol., 2017, 10.1113/JP274581 Glucagon, the body’s principal hyperglycaemic hormone, is released from α‐cells of the pancreatic islet. Secretion of this hormone is dysregulated in type 2 diabetes mellitus but the mechanisms controlling secretion are not well understood. In this study, we explore the importance of one such candidate mechanism by using an optogenetic strategy.
Variational Inference over Non-differentiable Cardiac Simulators using Bayesian Optimization A. McCarhty, B. Rodriguez, A. Minchole ArXiv, 2017, arXiv:1712.03353 The electrical signals that cause the heart to contract propagate through the torso and we can be recorded on an electrocardiogram (ECG). Cardiac simulators replicate this propagation. We develop a method to infer parameters that improve the fit of a state-of-the-art cardiac simulator.
High-throughput Binding Affinity Calculations at Extreme Scales J. Dakka, K. Farkas-Pall, D. W. Wright, S. J. Zasada, V. Balasubramanian, S. Wan, P. V. Coveney, S. Jha, ArXiv, 2017, arxiv.org/abs/1712.09168 Resistance to chemotherapy and molecularly targeted therapies is a major factor in limiting the effectiveness of cancer treatment. In many cases, resistance can be linked to molecular interactions between target proteins and the drug. Using multi-stage pipelines of molecular simulations, we can gain insights into the binding free energy between the two and the residence time for a drug, which can inform both stratified and personal treatment regimes and drug development.
Parameter estimation of platelets deposition: Approximate Bayesian computation with high performance computing R. Dutta, B. Chopard, J. Lätt, F. Dubois, K. Boudjeltia, A. Mira, arXiv, 2017, arxiv.org/abs/1710.01054 We propose a methodology which combines clinical images of platelet deposition in the blood, a mathematical model of the deposition process, and a HPC machine-learning approach which calibrates the model parameters to match the clinical images. The combined approach allows us to determine the deposition rates of platelets and detect anomalies, and, in turn, to contribute to a new generation of personalized platelet functionality tests.
Computational techniques for ECG analysis and interpretation in light of their contribution to medical advances A. Lyon, A. Michole, J.P.Martinez, P.Laguna, B. Rodriguez, Journal of the Royal Society Interface, 15 (138)2017, 10.1098/rsif.2017.0821 This is a review of the computational techniques that have been proposed for ECG analysis and computer simulations.
In silico evaluation of arrhythmia X. Zhou, A. Bueno-Orovio, B-Rodriguez, Current Opinion in Physiology, 1, 95-103, 2017, 10.1016/j.cophys.2017.11.003 This study highlights key studies in the field of computational simulations of arrhythmia
Host genotype and time dependent antigen presentation of viral peptides: predictions from theory R. Eccleston, P. V. Coveney, and N. Dalchau, Scientific Reports, 7 (1), 14367, 2017, 10.1038/s41598-017-14415-8 We construct a model of the recognition of HIV infection by the MHC (Major Histocompatibility Complex) class I pathway. The model is predictive and useful for helping to design experiments that provide a mechanistic understanding of immune recognition.
Phenotypic variability in LQT3 human induced pluripotent stem cell-derived cardiomyocytes and their response to antiarrhythmic pharmacologic therapy: An in silico approach M. Paci, E. Passini, S. Severi, J. Hyttinen, B. Rodriguez, Heart Rhythm, 2017, 10.1016/j.hrthm.2017.07.026 The study demonstrates, through the simulation of cell populations, the effect mutations in the electrophysiology of stem cell derived cardiomyocytes.
β-adrenergic receptor stimulation inhibits proarrhythmic alternans in post-infarction border zone cardiomyocytes: a computational analysis MJ. Tomek, B. Rodriguez, G. Bub, J. Heijman, American Journal of Physiology – Heart and Circulatory Physiology, 2017, 10.1152/ajpheart.00094.2017 The study demonstrates the anti-arrhythmic effects of beta-adrenergic stimulation post-myocardial infarction using detailed computer simulations.
Cellular Level In-silico Modeling of Blood Rheology with An Improved Material Model for Red Blood Cells G. Závodszky, B. van Rooij, V. Azizi and A. Hoekstra, Front. Physiol., 2017, 10.3389/fphys.2017.00563 A detailed model of blood flow based on modelling every single cell in a cubic millimeter of blood and validation against experimental data.
Multiscale Computing in the Exascale Era S. Alowayyed, D. Groen, P. V. Coveney, A. G. Hoekstra, Journal of Computational Science, 2017, 10.1016/j.jocs.2017.07.004 A vision on how to execute simulations that span multiple scale on the most powerful supercomputers that exist today, and those that will be available in the near future.
Rapid and accurate assessment of GPCR–ligand interactions Using the fragment molecular orbital-based density-functional tight-binding method I. Morao, D. G. Fedorov, R. Robinson, M. Southey, A. Townsend-Nicholson, M. J. Bodkin, A. Heifetz, Journal of Computational Science, 2017, 10.1002/jcc.24850 This paper explores drug interaction with proteins, G protein-coupled receptors, that are a common target for pharmaceuticals. With the particular computer simulation method harnessed in this work, the computational cost is decreased 1000-fold. By exploring and developing such computational methods, we move closer to a future where drug design is regularly informed by simulations.
The role of multiscale protein dynamics in antigen presentation and T lymphocyte recognition R. C. Eccleston, S. Wan, N. Dalchau, P. V. Coveney, Frontiers in Immunology, 2017, 10.3389/fimmu.2017.00797 We advocate a mechanistic description of antigen presentation and TCR (T-cell receptor), which involves multiscale modelling approaches collectively span several length and time scales. The approaches are capable of furnishing reliable biological descriptions that are difficult for experimentalists to provide.
A Comparison of Fully-Coupled 3D In-Stent Restenosis Simulations to In-vivo Data P. S. Zun, T. Anikina, A. Svitenkov, A. G. Hoekstra, Frontiers in Physiology, 8, 1-12, 2017, 10.3389/fphys.2017.00284 A model for scar tissue in coronary arteries after treatment of stenosis compared to data from animal experiments.
Exact solutions to the fractional time-space Bloch–Torrey equation for magnetic resonance imaging A. Bueno-Orovio, K. Burrage, Commun Nonlinear Sci Numer Simulat., 52, 91-109, 2017, 10.1016/j.cnsns.2017.04.013 This study demonstrate the exact solutions to the fractional time-space Bloch-Torrey equation for application in magnetic resonance imaging.
An Ensemble-Based Protocol for the Computational Prediction of Helix-Helix Interactions in G Protein-Coupled Receptors using Coarse-Grained Molecular Dynamics N. Altwaijry, M. Baron, D. Wright, P. V. Coveney, A. Townsend-Nicholson, Journal of Chemical Theory & Computation, 13 (5), 2254-2270, 2017, 10.1021/acs.jctc.6b01246 We provide a systematic, reproducible, and reliable protocol for determining the specific points of interaction between GPCR dimers. Our method is of great utility in further understanding GPCR function and also has broad applicability to many different types of membrane proteins.
Evaluation and Characterization of Trk Kinase Inhibitors for the Treatment of Pain: Reliable Binding Affinity Predictions from Theory and Computation S. Wan, A. Bhati, S. Skerratt, K. Omoto, V. Shanmugasundaram, S. Bagal, P. V. Coveney, Journal of Chemical Information and Modelling, 57 (4), 897-909, 2017, 10.1021/acs.jcim.6b00780 The paper presents free energy methods which could be used as tools to guide lead optimization efforts across multiple prospective structurally enabled programs in the drug discovery setting for a wide range of compounds and targets.
Opinion: Is big data just big hype? P. V. Coveney and R. Highfield, Longevity Bulletin: Big data in health, Institute and Faculty of Actuaries, 11-12, 2017, ISSN 2397-7213 To effectively use the explosion in big data, we need to understand the characteristics and sensitivity of the complex systems. The paper emphasises the importance of tools and models for pattern extraction and visualization, which need to be truly predictive.
Rapid and Reliable Binding Affinity Prediction of Bromodomain Inhibitors: a Computational Study S. Wan, A. P. Bhati, S. J. Zasada, I. Wall, D. Green, P. Bamborough, and P. V. Coveney, J. Chem. Theory Comput., 13 (2), 784-795, 2017, 10.1021/acs.jctc.6b00794 In collaboration with GlaxoSmithKline, we apply our ensemble based free energy approaches to accurately rank ligands by their binding free energies. The approach offers a long awaited development in the field of structure-based drug design.
Functional identification of islet cell types by electrophysiological fingerprinting L. J. B. Briant, Q. Zhang, E. Vergari, J. A. Kellard, B. Rodriguez, F. M. Ashcroft, P. Rorsman, Journal of Royal Society Interface, 14 (128), 1-20, 2017, DOI: 10.1098/rsif.2016.0999 This study shows the identification of islet cell types using electrophysiological information.
Atrial Fibrillation Dynamics and Ionic Block Effects in Six Heterogeneous Human 3D Virtual Atria with Distinct Repolarization Dynamics C. Shanchez, A. Bueno-Orovio, E. Pueyo, B. Rodriguez, Front. Bioeng. Biotechnol., 5, 1-13, 2017, DOI: 10.3389/fbioe.2017.00029 The study demonstrates HPC simulations of differences in atrial arrhythmias using six heterogeneous human three-dimensional models of atrial electrophysiology
Rapid, accurate, precise and reliable relative free energy prediction using ensemble based thermodynamic integration A. Bhati, S. Wan, D. Wright, P. V. Coveney, Journal of Chemical Theory and Computation, 13 (1), 210-222, 2017, 10.1021/acs.jctc.6b00979 We apply a systematic protocol for uncertainty quantification to a number of popular free energy methods. With a reliable measure of error estimation, ensemble-based simulations can be used to predict relative free energies accurately.

Acknowledgements

Dissemination (including electronic) of all results must:

  • Display the EU emblem
  • Include the following text for anything published up to September 2019: “This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 675451 (CompBioMed project)”
  • Include the following text for anything published after October 2019: “This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 823712 (CompBioMed2 project)”

Any dissemination of results must indicate that it reflects only the author’s view and that the Commission is not responsible for any use that may be made of the information it contains