Expanded Programme 26th September

Computer simulations are an indispensable tool in the drug development process to predict binding free energies, solubilities, and membrane permeability of drug candidates. Although the targets are very dierent, all free energy simulations share some common features and challenges. The two fundamental prerequisites for the determination of free energies are the accurate description of inter- and intramolecular interactions, and the adequate sampling of all relevant microstates. These two requirements are in conflict with each other, since a more sophisticated description of molecular interactions entails an increase of the computational costs, which inhibits the capability to search through a multitude of dierent possible conformations. In terms of the balance between those two requirements, one can distinguish two major classes of computational methods: a) classical force elds based on molecular mechanics (MM), which are fast and well suited for sampling, but involve approximations that limit their reliability b) quantum-mechanical methods (QM), which are based on molecular orbital calculations and combine a heavy computational burden with highly accurate interaction strengths. Full Abstract

Fragment-based lead generation (FBLG) involves scanning a library of low molecular weight compounds (fragments) to see if they bind to the target of interest and using those that do as building blocks to create create higher affinity molecules. A frequent strategy is to link multiple fragments binding different regions of the protein. FBLG represents an attractive application for in silico binding anity calculations, but the need to obtain comparable values from different binding modes represents a considerable challenge for many computational techniques. We evaluate the performance of our range of ensemble simulation based binding free energy calculation protocols, called ESMACS (enhanced sampling of molecular dynamics with approximation of continuum solvent). In studies of drugs binding to a single site these protocols have been shown to produce results which correlate well with experiment (correlation coecients >0.7) and provide robust uncertainty estimates. Full Abstract

The evolution of resistance to antibiotics was predicted by Fleming in his Nobel Prize speech and is now accepted as posing a threat to modern medicine requiring urgent and concerted action. Helping clinicians make appropriate treatment decisions by improving the coverage, portability, speed, accuracy and cost of species identification and drug susceptibility testing will be an important part of the solution. A promising approach is to sequence the genome of any infecting pathogen(s) found in a clinical sample and, by looking up genetic variants found in genes known to confer resistance to the action of antibiotics, return a prediction of the effectiveness, or otherwise, of a panel of antibiotics to the clinician. The exemplar for this approach is tuberculosis, partly because its growth rate is so slow that culture-based clinical microbiology can take up to two months to return a result to the clinician, and partly because its genetics is simpler than other pathogens and therefore the current second-generation sequencing technologies work well. Genetics clinical microbiology has been shown to be cheaper, faster and probably more accurate than traditional culture-based clinical microbiology
for the drug susceptibility testing of tuberculosis1 and, in addition, facilitates the rapid identification of to phenotype have been carefully and extensively developed, a potential weakness remains: such an approach is fundamentally inferential and so cannot make a prediction when it encounters a genetic variant not present in the catalogue, such as is the case for rare genetic mutations. Full Abstract

Free energy calculations have become a powerful addition to the computational chemist’s toolbox to support structure-based drug design in hit-to-lead and lead optimization stages of drug discovery projects. Methodological advances, the availability of less expensive large computational resources and automated workflows have opened up the possibility to apply the technology in an industry context at large scale. In 2016, we started a large initiative at Merck KGaA to thoroughly investigate the potential of free energy calculations for compound optimization and to define best practices for using this technology. Here, we present prospective data from using FEP+ in 10 drug active discovery projects at Merck KGaA over the course of three years and compare this performance to results obtained on a new, challenging benchmark of five pharmaceutically relevant targets. We further discuss opportunities and challenges and highlight use cases and conditions that can maximize the impact of the method. Full Abstract

Pore-forming protein analogues have been fabricated from triethylene glycol-cholesterol modified DNA sequences, which hybridize to form cholesterol anchored DNA nanopores (TEG-C NP’s). These versatile nanopores can be chemically tuned to exhibit an array of functionalities with a broad range of potential applications in biomedicine e.g. novel ligand controlled and light-controlled drug delivery systems[1,2,3]. The interactions between TEG-C NP’s and membrane lipids are pivotal to their function, but these interactions remain poorly understood. Here we use an ensemble-based, coarse-grained molecular dynamics (CG-MD) protocol to gather detailed, reproducible data on the structure and dynamics of TEG-C NP’s at two experimentally relevant ionic concentrations, allowing us to calculate reliable pore dimensions and perform comprehensive fluctuation analyses on membrane-spanning TEG-C NP’s, as well as TEG-C NP’s in free solution. Thus we can confidently characterise the influence of ionic concentration and membrane encapsulation on the dimensions, structural and mechanical properties of TEG-C NP’s, and pinpoint areas of constriction, strain and stability within their structure. Collecting ensembles of micro-second long trajectories of a membranespanning TEG-C NP allows us to observe a comprehensive spread of large-scale motions available to the TEG-C NP at these timescales and draw parallels with what is observed in experiment. Full Abstract

Water plays an intimate role in protein-ligand binding, not only through solvation/desolvation effects, but more subtly through the formation of direct interactions between the protein and ligand in the binding site. The targeting of bound water molecules for displacement as part of ligand optimization is a long invoked paradigm based around the release of configurational entropy, but there are many examples where displacing water leads to a loss in ligand binding affinity. Quantitatively accurate approaches to address this problem are arguable inadequate – water displacement and ligand interactions are intimately related and difficult to disentangle both experimentally and, hitherto, computationally. We have a long-standing interest in developing and using Grand Canonical Monte Carlo (GCMC) simulation approaches to explore water binding in protein-ligand systems. Through GCMC we are able to locate water molecules with good accuracy when compared against crystal structures. More significantly, the simulations clearly demonstrate the important role of water cooperativity; the mutual tabilization of water molecules means that individual water molecules cannot always be considered in isolation, but rather as part of a network. GCMC allows water binding sites and network binding free energies to be simultaneously calculated. In addition, by combining GCMC with alchemical perturbations of the ligand, networks of bound water molecules are able to adapt and maintain equilibrium with bulk water as the perturbation proceeds. Furthermore, the ability to extract active-site hydration free energies allows the deconvolution of protein-ligand binding free energies into separate protein- and water-mediated components, thereby providing rich, additional detail to the structure-activity relationship (SAR). In this presentation, the underlying methodology GCMC methodology will be described, together with examples of its application to water placement, binding free energy calculations, and protein-ligand affinity prediction Full Abstract

This talk will review some of the challenges posed by the huge growth of experimental data generated by the new generation of large-scale experiments at UK national facilities at the Rutherford Appleton Laboratory site at Harwell near Oxford. Such ‘Big Scientific Data’ comes from the Diamond Light Source and Electron Microscopy Facilities, the ISIS Neutron and Muon Facility, and the UK’s Central Laser Facility. Increasingly, scientists are now needing to use advanced machine learning and other AI technologies both to automate parts of the data pipeline and also to help find new scientific discoveries in the analysis of their data. For commercially important applications, such as object recognition, natural language processing and automatic translation, Deep Learning has made dramatic breakthroughs. Google’s DeepMind has now also used Deep Learning technology to develop their AlphaFold tool to make predictions for protein folding. Remarkably, they have been able to achieve some spectacular results for this specific scientific problem. Could Deep Learning be similarly transformative for other scientific problems? After a brief review of some initial applications of machine learning at the Rutherford Appleton Laboratory, we focus on challenges and opportunities for AI in advancing pharmaceutical and materials science. Finally, we discuss the importance of developing some realistic machine learning benchmarks using Big Scientific Data coming from a number of different scientific domains. For the computer vision community, it was the ImageNet database that provided researchers with the capability to evaluate algorithms for object detection and image classification at large scale. The ImageNet Large Scale Visual Recognition Challenge (ILSVRC) allowed researchers to compare progress in detection across a wider variety of objects and led directly to the present Deep Learning and GPU revolution. We believe that the creation of a credible ‘Scientific Machine Learning’ (SciML) collection of benchmarks could prove useful and significant for the scientific research community. The talk concludes with some initial examples of our ‘SciML’ benchmark suite and a discussion of the research challenges these benchmarks will enable. Full Abstract

Artificial intelligence is underway to transform the society through technologies like self-driving cars. Also, in drug discovery machine learning and artificial intelligence methods has received increased attention. [1] The increased attention is not only due to methodological progress in machine learning and artificial intelligence, but also progress in automation for screening, chemistry, imaging and -omics technologies, which have generated very large datasets suitable for machine learning.
While machine learning has been used for a long time in drug design, there has been two exiting developments during the last years. One is the progress in synthesis prediction, where deep learning together with fast search methods like Monte Carlo Tree Search has been shown to improve synthetic route prediction as exemplified by a recent Nature article. [2] In this talk I will focus on the second development, which is applying deep learning based methods for de novo molecular design. It has always been the dream of the medicinal and computational chemist to be able to search the whole chemical space of estimated 1060 molecules. This would be a step change compared to search enumerable chemical libraries of perhaps 1010 compounds. Methods to search the whole chemical space through generative deep learning architectures has been developed during the last 3-years. In the presentation there will be a focus de novo generation of molecules with the Recurrent Neural Network (RNN) architecture. The basis will be described and exemplified of how molecules are generated. After the concept has been introduced it will be described how the method is used within drug design projects at AstraZeneca. Current limitations will be discussed in conjunction with mitigation strategies to further enhance the potential of RNN based molecular de novo generation. Full Abstract

Due to recent advancements in Deep Learning (DL) algorithms and frameworks, we have started to witness the convergence of High Performance Computing (HPC), Machine
Learning (ML), and various application domains, such as healthcare. This opens the possibility to address the high complexity problems that deal with large data and were considered unsolvable in the past. In this talk we will present several use-cases going from synthetic to real-world problems for medical image classification, segmentation, and generation, using both 2-D and 3-D data. The focus will be on the scale-out behavior and best practices, while also giving details into the bottlenecks encountered in the various use-cases. Jointly working within Intel’s IPCC (Intel Parallel Computing Centers) program, we will present SURFsara’s collaborations with DellEMC, NKI (Netherlands Cancer Institute), and the EXAMODE (www.examode.eu) project consortium. We will demonstrate how large memory HPC systems enable solving medical AI tasks. Full Abstract

The Cancer Deep Learning Environment (CANDLE) is an open framework for rapid development, prototyping, and scaling deep learning applications on high-performance computing (HPC) systems. CANDLE was initially developed to support a focused set of three pilot applications jointly developed by cancer researchers and deep learning / HPC experts, but is now generalizable to a wide range of use cases. It is designed to ease or automate several aspects of the deep learning applications development process. CANDLE runs on systems from individual laptops to OLCF Summit, the most powerful supercomputer in the world, and enables researchers to scale application workflows to the largest possible scale. Full Abstract

Effective modelling across the genomic scales within a cellular environment plays a crucial role in understanding the principles that govern cell cycle aberration, for instance cancer or disease. The selection of alleles, in conjunction with RNA and protein concentrations, with epigenetic factors; contribute significantly to the cell state and capacity to function. Further to this, sequence-derived features (SDFs) derived from DNA, RNA and protein sequences can contribute useful static information in conjunction with these dynamic processes to improve inference and control for steady-state effects in measurement data. These are commonly applied in transcriptomic studies whereby mRNA level acts as a proxy for protein abundance, as SDFs can be added to the model to improve predictive power. A major limiting factor of many previous studies has been lack of supportive data to coincide expression levels in the analysis of various biological domains. Full Abstract

The explosion of healthcare information stored in Electronic Health Records (EHR) has led to an increasing trend of EHR-based applications in computational biomedicine. Unfortunately, applying deep learning (DL) to medicine is no trivial task as EHR data is extremely complex, usually unbalanced, muddled with missing or invalid values and frequently contains a heterogeneous mixture of data types and structured/unstructured formats. The problem has been compounded by the lack of publicly available datasets that are large enough for the development of deep learning methods as well as by the lack of benchmarking tasks and metrics to compare results. The creation of the MIMIC-III Clinical Database [1] and the recent work of Harutyunyan et al. [2] proposing benchmarking tasks and metrics are accelerating advances in the field. Full Abstract

Recently, generative models based on deep neural networks have been proposed to perform de-novo design, that is to directly generate molecules with required property profiles by virtual design-make-test cycles [1,2]. Neural generative models can learn to produce diverse and synthesisable molecules from large datasets, for example by employing recurrent neural networks, which makes them simpler to set up and potentially more powerful than established de novo design approaches relying on hand-coded rules or fragmentation schemes. Full Abstract

Marketing approvals of (new) medicinal products (and combinations) generates large interests of both patients in need of new medicinal therapies and the sponsors (big pharmaceutical industry, SMEs, and academia). Before a new medical product can be used on humans in a country, it must be approved for that use by the relative Regulatory Authority of that country. In USA, this will be the remit of the Food and Drug Administration (FDA) drugs and medical devices whereas in the European Union pharmaceuticals are approved by the European Medicine Agency (EMA) mostly and by national competent authorities to a lesser extent. The approval of medical devices is delegated to the member states, through selected notified bodies.
Consequently, regulators have to find the appropriate balance between the need to ensure that decision-making is based on scientifically valid data and the need for access to the new medicines is considered.
Full Abstract

Coronary Artery Disease (CAD) is the leading cause of death in Europe and worldwide with more than 17 million deaths [1]. Atherosclerosis, the major disease process of CAD, is a chronic inflammation driven by the build-up of atherosclerotic plaques inside the coronary arteries. Bioresorbable Vascular Scaffolds (BVS) revolutionised the field of interventional cardiology by providing targeted drug delivery, mechanical support and complete resorption overcoming the barriers of bare-metal and drug-eluting stents. In vitro and in vivo experiments followed by clinical trials are currently used in providing useful information on the safety and efficacy of BVS. However, these processes are time-consuming and costly. In parallel, they raise ethical considerations due to the uncertainty related to the extremely well performance in controlled laboratory experiments and pre-clinical studies, and potential under performance during or after clinical trials. Full Abstract

In-Silico Trials (IST) represent an innovative application of Virtual Human technology helpful in assisting and supporting the refinement, the reduction, or the replacement of pre-clinical and clinical trials. In this multifaceted challenge perspective, the regulatory authorities are facing with an increasing number of projects developing and applying ISTs ranging from validating underway in-silico models of specific pathophysiology or applied virtual populations, via technological and infrastructural demands. The last few years have been characterised by an intense activity around the so-called regulatory science, aimed to ensure a robust approach to assess the credibility of individual in-silico methods as sources of regulatory evidence (Viceconti et al., 2017; Pappalardo et al., 2018a; Morrison et al., 2018). Full Abstract

The clinical assessment of new drugs against osteoporosis is a particularly expensive one. Ideally a clinical trial should have fragility fractures as primary outcome, and should follow-up patients for at least five years, but this would bring the cost and time-to-market to unacceptable levels. A quick search on https://clinicaltrials.gov shows that Romosozumab, one of the latest drugs brought to market has been studies either with indirect outcome metrics (such as bone density), or at most observing fractures over 24 months. While this is probably adequate for the specific purposes of those trials, the possibility to assess the efficacy of these new drugs with in silico trials is of extreme interest. Full Abstract

Familial cardiomyopathies (FCM) are most commonly diagnosed, or progress of the disease is monitored, through in vivo imaging, with either echocardiography or, increasingly, cardiac magnetic resonance imaging (MRI). The treatment of symptoms of FCM by established therapies could only in part improve the outcome, but novel therapies need to be developed to affect the disease process and time course more fundamentally. In SILICOFCM project we are doing in silico multiscale modeling of FCMs that would take into consideration comprehensive list of patient specific features (genetic, biological, pharmacologic, clinical, imaging and patient specific cellular aspects) capable of optimizing and testing medical treatment strategy with the purpose of maximizing positive therapeutic outcome, avoiding adverse effects, avoiding drug interactions, preventing sudden cardiac death, shortening time between the drug treatment commencement and the desired result. Full Abstract

The incorporation of computational approaches into the early drug design process is a relatively young discipline compared to the long-standing history of drug discovery research. Considerable advances in hardware architecture, speed, accuracy and usability of computational algorithms have paved the way towards a quickly developing branch of research within the pharmaceutical industry. This talk will provide a general overview of computational molecular design in a pharmaceutical industry setting, highlight recent methodological advances and discuss their impact on real-world drug discovery projects.
Full Abstract

There is an urgent need in the pharmaceutical industry for approaches and tools that are able to accurately, rapidly predict binding affinity values. Previous work[1-2] has demonstrated the inability of ‘one-off’ simulations to accurately, reliably and reproducibly predict the overall conformational states and dynamics of biological systems over a finite period of time. Thus, the use of enhanced sampling techniques is essential for accurate descriptions of binding between a receptor and its ligands. We investigate the application of Enhanced Sampling of Molecular Dynamics with Approximation of Continuum Solvent (ESMACS), and Thermodynamic Integration with Enhanced Sampling (TIES) for computing the binding affinities (see Figure 1) of a series of experimentally verifiable ligands to the A1 and A2A adenosine receptors (see Figure 2), members of a subclass of the GPCR superfamily. Full Abstract

Drug-target residence time, the lifetime of the ligand-receptor complex, is said to be better than binding affinity at predicting in vivo efficacy. Computational prediction of drug-target residence time, using standard molecular dynamics (MD), is challenging as experimental dissociation times are approximately 107 longer than the simulation times that are currently feasible. We therefore applied steered MD (SMD) to forcibly speed up ligand dissociation. To ensure the interactions of the dissociating ligand with the receptor residues and water, Figure 1, are reproducible, ensemble analysis was performed. We applied this method to 17 ligands of a prototypical GPCR (G protein-coupled receptor), all of which had associated published experimental kinetic binding data. Our results reveal that the computationally-calculated change in ligand-water interaction energy correlates strongly with experimentally-determined residence time (R2 = 0.79). Further, the residues that interact with the dissociating ligand in these simulations are known experimentally to affect binding affinity and residence time. These experimental data indicate that our ensemble-based SMD protocol[1] is a novel, rapid and reproducible method for the rationalisation and determination of drug-target relative residence time. Full Abstract

G-protein coupled receptors (GPCRs) constitute the most important drug target family and account for 30% of the FDA approved drugs1. This large family of receptors detect a remarkably diverse array of molecules outside the cell and initiate a variety of intracellular signalling pathways in response. The transmembrane nature and intrinsic flexibility of GPCRs makes their crystallization difficult. But a number of technical advances, aiming to rigidify the receptor have allowed their crystallization increasing the number of available structures. Despite this breakthrough in crystallography, which lead to the Nobel prize in chemistry to Lefkowitz and Kobilka in 20122,3, these structures are unlikely to cover the conformational diversity of this family of receptors and must be complemented with other techniques to reveal the intrinsic dynamics of the process. We are only starting to understand the role of ligand induced conformational changes (allostery) in GPCRs and there remains a great deal to be discovered in order to facilitate fundamental understanding of the role of allostery and the potential of new allosteric drugs4,5. Here we present a combination of state-of-the-art molecular dynamics enhanced sampling techniques and force fields to understand at an atomistic level how ligands and intracellular partners affect the energy and interconversion rates of GPCRs conformational repertoire. Our study is focused on a prototypical class A GPCR, the adenosine receptor A2a, which is relevant to the occurrence, development and treatment of brain ischemic damage and degenerative disorders, due to its role as neuronal and synaptic function modulator. Full Abstract

Retinoids are a class of vitamin-A derived molecules with endogenous roles in cell proliferation and differentiation. Recent research has suggested retinoids may hold promise for therapeutic use in motor neuron diseases such as amyotrophic lateral sclerosis (ALS) by promotion of neuronal survival. Despite promising therapeutic potential, little is known about the complex signalling pathways which govern retinoid’s mechanism of action. Endogenous retinoids such as all-trans-retinoic acid (ATRA) are inherently vulnerable to photodegradation and isomerism due to their polyene structure, making their use as a research tool problematic. As such we utilise a range photostable, fluorescent synthetic retinoid analogues we are currently developing between Durham University and LightOx to investigate the retinoid mode of action. In parallel with biological testing, ligand docking and molecular dynamics (MD) simulations form a vital part of our continued research into these compounds. As part of our docking analysis, we have developed a root-mean-square deviation (RMSD)-based clustering script to group and identify commonly occurring ligand-docked protein structures, which we hypothesise will allow for enhanced identification of promising docked solutions via less resource-intensive methods before moving data to resource-intensive MD simulations. This workflow not only introduces a new approach for docking analysis but allows for faster and simpler identification of unique protein-ligand docked solutions. Full Abstract

Osteoporosis is a bone disease characterised by brittle bone and increased fracture incidence. The disease is globally a high burden on health systems which continues to increase with an aging society. There are limited treatments for osteoporosis with just two FDA approved pharmacological agents in the USA. Furthermore the drug discovery pipeline has limited success in producing novel and efficacious molecules. Osteoporosis arises due to changes in bone architecture, mineral density (BMD) and strength. These characteristics are believed to be affected by bio-mechanical stimulations. Such signals are sensed by bone cells in the bone remodelling unit and translated to cellular responses which ultimately maintain healthy bone. Recently, dual bio-mechanical stimulation with intermittent parathyroid hormone (PTH) treatment and mechanical stimulation were shown to increase BMD and bone formation in mice, thus suggesting a promising treatment for osteoporosis 1,2. However, the exact regimes to induce potent therapeutic effects are yet uncharacterised. This is partly due to incomplete understanding of cellular and molecular mechanisms which sense and integrate the dual signals into a cellular response (i.e. mechanotransduction) which evolve into increased BMD, strength and growth at the tissue level. Full Abstract

Binding free energy calculations (BFE’s) are routinely used in drug design to accurately predict the binding free energy of small molecules to drug targets,1 however the cost of simulation often prohibits their application to smaller sets of molecules. Groups of molecules are typically compared through free energy maps, where each ligand may be compared to at least two other small molecules, however the decision process involved in the generation of this map is un-rigorous. Certain calculations between pairs of ligands or even a given atom-mapping protocol will converge faster than others, proportional to the thermodynamic length of the specific transformations. Using perturbations with large thermodynamic lengths is inefficient,2 however it is not possible to calculate thermodynamic length a priori, and can only be established post simulation.3 Typically, the generation of a ligand free energy map involves naïve reasoning over which pairs of ligands to compare, while the efficiency of the chosen pairings, and therefore the overall quality of the map is only apparent post simulation. Each perturbation is generally simulated using ‘equal allocation’ whereby the same length of simulation is used for each perturbation. Full Abstract

In Barcelona, July 2018, the Barcelona Supercomputing Center (BSC), the Technical University of Catalonia (UPC), the Spanish Superior Council of Scientific Research (CSIC), three researchers and a business-background entrepreneur, founded the start-up company ELEM Biotech (http://www.elem.bio). The company is in charge of developing and commercialize a set of high-tech simulation technology targeting the biomedical sector. The core technology is Alya, the parallel multi-scale / multi-physics simulation software developed at the Barcelona Supercomputing Center, by a team led by Mariano Vázquez and Guillaume Houzeaux. The company is based in Barcelona, with an office in Bristol (UK). Full Abstract

There are typically two complementary types of compute workload taking place in academic groups looking to spin-off into commercial entities:
1. Your known, production workflows which are showing promise for building up a company that can pioneer the way forward in science.
2. Your unknown, research concepts, ideas, and emerging workflows that can feed the cycle of improvement.
So how do you build up commercial compute without expending too much time and killing your budget in betting on the wrong resources?
For the past three years Alces Flight has been studying how public cloud is impacting High Performance Computing (HPC) as an emerging platform. Our exploration moved us to initially create a tool designed to launch an HPC environment straight in the cloud – but what we learned along the way has made us re-evaluate how emerging business can harness different resources in order to build a smart cycle of compute that can feed academics transitioning into company life. At the September, 2019 CompBioMed Conference we wish to share the insights made in:
• Honing characteristics of your workload to understand what constitutes production (paid commercial) work and what constitutes the research to feed improvements.
• Understanding how spreading your compute budget can help you not only survive getting started, but make strong future investments.
• Balancing both free and paid technologies and services to avoid the trap of believing you can do everything in-house. Full Abstract

Since the earliest virtual models of molecules and simulations, incredible efforts in technology and methodology have brought forth solutions and tools now employed in the understanding of biomolecular interactions
and prediction of their properties. The development and performance of these solutions were challenged by their diversity but have definitely lead to the increased use of computerized methods in a wide range of research fields, from genomics to drug design.
Acellera has acquired a strong expertise in software development and structural studies over the last 10 years. We designed, alone or in collaboration, innovative solutions for the understanding of critical events for molecular recognition like ligand binding and conformational changes of biomolecules, key steps in the drug design process. Complex protocols combined with the need for high performance infrastructure hampered the access and use of such solutions by the whole scientific community dedicated to Drug Discovery. Full Abstract

For decades, universities and research centers have been applying modeling and simulation (M&S) to medical devices [1] and pharmaceutical [2] development, coining the new expression in silico clinical trials. Its use however is still limited to a restricted pool of specialists.
Making M&S available to a broad spectrum of potential users (medical device and pharmaceutical companies, hospitals, healthcare institutions) would require an easy and controlled access to M&S resources in a secure environment. A joint effort between academia, industry and regulatory bodies is therefore needed to reach a rapid adoption of a harmonized approach.
It is here proposed an easy-to-use cloud-based platform that aims to create a collaborative marketplace for M&S in healthcare, where developers and models’ creators are able to capitalize on their work while protecting their intellectual property (IP), and medical device and pharmaceutical companies can use M&S to accelerate time and to reduce costs of their research and development (R&D) processes. Full Abstract

In my role as UCL lead for CompBioMed, a H2020 Centre of Excellence in Computational Biomedicine (compbiomed.eu), and as Head of Teaching for Molecular Biosciences at UCL, I have integrated research and teaching to lead the development of HPC-based education targeting medical students and undergraduate students studying biosciences in a way that has explicitly designed to be integrated into their existing university programmes as credit bearing courses. This innovation has not been replicated in any other university in the world. One version of the taught course has been designed for medical students in Years 1 and 2 of study (SSC334) and one of the unique features of the course is the integration of experimental and computational aspects, with students obtaining and processing biological samples, using state of the art Next Generation Sequencing and then interrogating the DNA sequences computationally using code that was ported to high performance computing (HPC) facilities of CompBioMed’s HPC Facility core partners (EPCC (UK), SURFsara (Netherlands) and the Barcelona Supercomputing Centre (Spain)). Another version of the taught course (BIOC0023) replaces the final year research project course for undergraduate biomedical science students, providing them with the opportunity to design and complete an entire research project from developing experimental hypotheses to investigating these in a way that involves the integration of experimental and computational methodologies. In the past 18 months, these UCL courses have successfully run with a total of ~250 students participating (60 medical students and 195 biomedical science students).
Full Abstract

The realisation that biological processes can be explained in terms of the interactions between a limited set of fundamental chemical building blocks has driven progress in biomedical science for more than a century, and has resulted in the molecular biology revolution. The key interdisciplinary boundary during this period was the frontier between medicine and chemistry. Today, however, a new set of challenges is changing the frontier of biomedicine. Technological advances in every area of biology have resulted in an exponential increase in the rate of data acquisition, and in its complexity. There is a growing consensus that the rate at which we are acquiring new hyper-dimensional biological data now outstrips our ability to analyse, integrate, interpret and eventually exploit it to drive progress in medicine and improve health. In order to tackle these challenges, biomedical scientists need new sets of skills. Specifically, mathematics offers a framework with the potential to simplify the increasingly complex data which is being produced, and provide fundamental rules which capture the behaviour of physiological systems in health and disease, and allow us to predict how they may respond to different types of therapeutic intervention. In parallel, artificial intelligence provides a means to implement this mathematics, using the increasingly sophisticated algorithms of machine learning. Full Abstract

Involving undergraduate students in research has proven to be an essential experience that enhances the learning outcome of students [1,2]. When exposed to research world at an early stage of their career, they acquire new skills that also guide them in tailoring their experience and choosing their future career. The High Energy and Medical Physics Group at Texas A&M University at Qatar (TAMUQ) has been supporting and engaging undergraduate students in different research projects in the areas of High Energy Physics and Medical Physics for the past six years. It attracted students from all four majors offered at TAMUQ, Electrical, Mechanical, Chemical and Petroleum Engineering shortly after the launching of its research activities. Many projects conducted within the group were awarded by the Qatar National Research Fund which is a governmental funding body that provides funding to highly competitive projects that address national priorities and contribute to capacity building [3]. Students were trained to use a high performance computing facility, different programming languages, software and Monte Carlo based platforms for their simulation. As for the outcomes, they participated and presented at international conferences, many of them attended CERN summer internship program. Some of the published journals in international peer-reviewed scientific journals.
In this paper, we present some of the projects that our students completed, the different tools that were used, as well as the research outcomes. Then, we will discuss the impact of this experience on their learning and undergraduate education as well as their career path, especially their postgraduate studies. Full Abstract

In this talk, I will describe an emerging initiative at Argonne National Laboratory to advance the concept of Artificial Intelligence (AI) aimed at addressing challenge problems in science. We call this initiative “AI for Science”. The basic concept is threefold: (1) to identify those scientific problems where existing AI and machine learning methods can have an immediate impact (and organize teams and efforts to realize that impact); (2) identify areas of where new AI methods are needed to meet the unique needs of science research (frame the problems, develop test cases, and outline work needed to make progress); and (3) to develop the means to automate scientific experiments, observations, and data generation to accelerate the overall scientific enterprise. Science offers plenty of hard problems to motivate and drive AI research, from complex multimodal data analysis, to integration of symbolic and data intensive methods, to coupling large-scale simulation and machine learning to drive improved training to control and accelerate simulations. A major sub-theme is the idea of working toward the automation of scientific discovery through integration of machine learning (active learning and reinforcement learning) with simulation and automated high-throughput experimental laboratories. I will provide some examples of projects underway and lay out a set of long-term driver problems. Full Abstract

Protein misfolding is a fundamental disease process implicated in many human disorders, particularly dementias such as Alzheimer’s disease, but also in diabetes and specific types of heart and kidney failure. Proteins are the structural and functional workhouse molecules of the human body – beneficial activities that are dependent upon the protein being folded into a correct shape. Since protein shape is central to health, it is reasonable to postulate the existence of compounds endogenous to the human body which ameliorate the pathological consequences of protein misfolding; the concept of searching for endogenous anti-protein misfolding compounds is unique. We have used an in silico high throughput screen of small molecules endogenous in the human brain to identify multiple classes of agents capable of inhibiting the aberrant protein misfolding implicated in the pathogenesis of Alzheimer’s dementia. We then extend this discovery to show that these endogenous compounds also inhibit the misfolding of proteins implicated in diabetes, thereby demonstrating that these agents are not disease specific and are applicable to multiple classes of protein misfolding diseases, including two of the most significant disorders afflicting humankind, namely dementia and diabetes.Full Abstract

The human genome consists of approximately 3 billion base pairs, stored as nucleic acid sequences. Due to its vast complexity, the genome is fragile – unsurprisingly the DNA within is susceptible to change. The mutations that occur in these DNA sequences are crucial to both natural evolution and the occurrence of genetic diseases. While some of these changes might be a consequence of exposure to high energy electromagnetic fields or other forms of radiation, mutations may also arise due to mistakes during the DNA replication process.
Although remarkably accurate, the high-fidelity DNA replication process generates base substitution errors at a rate of 10-4 to 10-5 per replicated nucleotide. However, due to >various intrinsic repair mechanisms, errors in human genome replication are actually less frequent (approx. ~10-8 to 10-10 per replicated nucleotide). These replication errors, known as p>oint mutations, may occur as a result of wobble base pairing, Hoogsteen (anti-syn) base pairing, ionisation and tautomerisation (the frequency of each is uncertain). Full Abstract

Clinical beam accelerators are widely used in radiation therapy facilities to provide the adequate beams for treatment. The success of the treatment depends on the accuracy of the dose calculated and radiation administered. In this work we conduct a comprehensive modelling of the Varian Clinac 2300C/D from electron beam generation to target response, using GATE simulation toolkit. To validate our numerical model, Gamma Index parameter is used to compare the simulation results against the experimental measurements. Results from Percent Depth Dose and Dose Profile are presented and discussed. This Monte Carlo model and the accuracy of these results can be extended to accurately calculate the dose distribution in real treatment planning systems. Full Abstract

Tight junctions are cell-cell contact structures found in epithelial and endothelial tissues, located at the contact region between neighbouring cells, towards their apical side. They regulate the permeability of small molecules and ions through the intercellular space (paracellular pathway) by either blocking their passage or allowing some molecules with appropriate charge and size to go through. A functional tight junction barrier is critical to the physiology of the body. Its dysregulation can lead to pathologies such as inflammation, metastasis and edema [1]. For instance, mice that lack a key tight junction protein, claudin 1, die after birth due to excessive water loss across the skin. Tight junctions are also the target of several viruses including the hepatitis C virus and bacteria such as the bacterium Clostridium perfringens that produces the enterotoxin responsible for food poisoning. They are also targets of strategies for enhancing drug delivery of large molecules including proteins across the gastrointestinal tract. Further, there are numerous hereditary diseases that are linked with mutations of tight junction proteins, which include hypomagnesemia, deafness, neonatal sclerosing cholangitis with ichthyosis and familiar hypercholanemia.Full Abstract

In diverse cellular pathways including clathrin-mediated endocytosis (CME) and viral bud formation, cytosolic proteins must self-assemble and induce membrane deformation. To understand the mechanisms whereby assembly is triggered and how perturbations can lead to dysfunction requires dynamics of not just assembly components, but their coupling to active, force-responsive, and ATP-consuming structures in cells. Current computational tools for studying self-assembly dynamics are not feasible for simulating cellular dynamics due to the slow time-scales and the dependence on energy-consuming events such as phosphorylation. We recently developed novel reaction-diffusion algorithms and software that enable detailed computer simulations of nonequilibrium self-assembly over long time-scales [1]. Our simulations of clathrin-coat assembly in CME reveal how the formation of structured lattices impacts the kinetics of assembly, and how localization to the membrane can stabilize large, dynamic assemblies not observed in solution. We developed a relatively simple equilibrium theory to quantify how localization of protein binding partners to the membrane can dramatically enhance binding through dimensionality reduction, providing a trigger for assembly [2]. Tuning the speed and success of vesicle formation can be sensitively controlled by the stoichiometry of assembly components, particularly those that control membrane localization through lipid binding [3]. Our results suggest that stoichiometric balance and membrane localization can act as potent regulators of self-assembly, and our reaction-diffusion software provides a powerful tool to characterize dynamics within the cell. Full Abstract

Clinical trials are expensive, and the risks posed to the participants are not fully known. Yet, participant numbers in clinical trials are often too small to conclude a statistically significant positive effect of the proposed intervention. A case in point are interventions targeting the reduction of hip fracture risk due to ageing in women. The socioeconomic relevance of this condition is well-known: for women over the age of 50, the remaining lifetime risk of suffering a hip fracture is equivalent to that of breast cancer; the cost of treating fragility fractures at the hip is over £2 billion annually in the UK. Yet, hip fracture incidence in the general population is very small (32 fractures per 10,000 person-years in British women over 50 [1]). This impedes reaching a statistically significant conclusion in a clinical trial with fracture as endpoint. In silico clinical trials (ISCTs) have been proposed as a computational tool to alleviate such challenges. Here, virtual patients are recruited in the trial increase the confidence in the study result. A virtual patient is a digitised data-set comprising biomedical information relevant to the disease/condition and treatment in question. An ISCT simulates a standard trial by subjecting virtual patients to untreated and treated conditions, where each condition is expressed by a mathematical model. Therefore, two ingredients are indispensable in any ISCT, irrespective of the intervention: a virtual patient definition and a mathematical model for the untreated condition. Full Abstract

Image-based subject-specific (SS) musculoskeletal models are currently considered the preferred solution in the estimation of biomechanical parameters due to their high accuracy and reliability compared to scaled generic models (GS) (Lenaerts et al., 2009). The high accuracy and reliability however come at the extra burden of obtaining medical images and time to build the models from these images (Valente et al 2014). The use of subject-specific models also becomes a challenge when analysis must be performed based on data for which imaging data is not available such as in a retrospective study and in most of standard clinical scenarios. Full Abstract

Despite significant progress in structure determination techniques such as X-ray crystallography/nuclear magnetic resonance (NMR), our ability to obtain atomistic insights into complex biological phenomena remains challenging. Understanding the relationship between conformational flexibility and function is an issue that experimental structural biology continues to struggle with and flexible regions in proteins are often absent from the structures obtained using X-ray crystallography and NMR techniques. Techniques such as neutron/Xray scattering and cryo-electron microscopy (cryo-EM) can potentially provide structural insight into flexible and heterogeneous biomolecular assemblies, but their inherent low resolution makes it difficult to elucidate atomistic/mechanistic details.
Although single particle cryo-EM is projected to displace traditional techniques for structure determination of large proteins and complexes (~150 kDa), flexible and heterogeneous biological assemblies present obvious challenges in approaching atomic resolution with this technique. On the other hand, small-angle scattering (SAS) with X-rays (SAXS) and neutrons (SANS) has been very informative for studying the structures of flexible systems such as multi-domain proteins and membrane proteins. When combined with computational modeling, it is possible to determine not only the ensemble of structures that reflect the conformational state of a protein but also the dynamical properties at relevant temporal scales of the system.Full Abstract

Major factors contributing to the acceleration of personalized healthcare in recent years come from advances in high-performance computing (HPC), data analytics, machine learning, and artificial intelligence, enabling scientists now to perform the most sophisticated simulations, in genomics, proteomics, and many other fields, using methods like genome analysis, molecular dynamics, and more general computer aided analysis methods widely applied and proven in other areas of scientific and engineering modelling.
To just select one, we demonstrate the impact of computer simulations on personalized health care and present a research project aiming at living heart simulations, which has recently been rewarded with several prestigious international awards. Full Abstract

In recent years it has become possible to calculate the binding affinities of compounds bound to proteins via rapid, accurate, precise and reproducible free energy calculations. The ability to do so has applications from drug discovery to personalised medicine. This approach is based on molecular dynamics simulations and draws on sequence and structural information of the protein and compound concerned. Free energies are determined by ensemble averages of many molecular dynamics replicas, each of which requires hundreds of CPU or GPU cores, typically requiring a supercomputer class resources to be calculated on a tractable timescale. In order to perform such calculations there are also requirements for initial model building and subsequent data analysis stages. Full Abstract

In this paper we present the outcome of three years of development work in the EurValve project [1] which resulted in the creation of an integrated solution for medical simulations referred to as the Model Execution Environment (MEE). Starting with a definition of the problem (which involves simulating valvular heart conditions and the outcomes of treatment procedures) we provide a description of the high-performance computational environment utilized to process retrospective use cases in order to create a knowledge base which underpins the EurValve Decision Support System (DSS). We also provide specific examples of MEE usage and the corresponding statistics. Full Abstract

is paper outlines an approach for enabling access to HPC applications as Software as a Service (SaaS) on conventional high-end HPC hosts such as EPCC’s Cirrus and cloud providers such as Microsoft’s Azure service. The focus for the approach is enabling access to the HemeLB application with the Polnet biomedical workflow. The paper reports on an implementation of this approach that allows Polnet workflows to run on the Cirrus and LISA supercomputing services at EPCC and SURFsara respectively. Full Abstract

Structural biology deals with the characterization of the structural (atomic coordinates) and dynamic (fluctuation of atomic coordinates over time) properties of biological macromolecules and adducts thereof. Gaining insight into 3D structures of biomolecules is highly relevant with numerous application in health and food sciences.
Since 2010, the WeNMR project (www.wenmr.eu) has implemented numerous web-based services to facilitate the use of advanced computational tools by researchers in the field, using the grid computational infrastructure provided by EGI. These services have been further developed in subsequent initiatives under the H2020 EGI-ENGAGE West-Life project and the BioExcel Center of Excellence for Biomolecular Computational Research (www.bioexcel.eu). The WeNMR services are currently operating under the European Open Science Cloud with the H2020 EOSC-Hub project (www.eosc-portal.eu), with the HADDOCK portal (haddock.science.uu.nl) sending >10 millions jobs and using ~2700 CPU years per year.
In my talk, I will summarize 10 years of successful use of e-infrastructure solutions to serve a large worldwide community of users (>13’500 to date), providing them with user-friendly, web-based solutions that allow to run complex workflows in structural biology. Full Abstract