CompBioMed All Hands Meeting 2020

The development of new drugs is a very expensive process that can be benefited from the use of structural bioinformatics techniques (SBT). Besides, application of SBT is very costly since it requires of demanding computations and this situation has reached a plateau in the last decades due to limitations not overcome yet. In this talk we present how SBT in drug discovery can be drastically accelerated with the use of high performance computing (HPC) architectures such as supercomputers and Graphics Processing Units (GPUs)¹ and methodological advancements that our group (http://bio-hpc.eu) has developed in the last years, and which have been successfully applied to the discovery of active compounds in the context of anticoagulants², cancer³, Parkinson⁴, Fabry⁵, nutraceuticals⁶, Zika⁷, etc. Finally, we also show how other users can access our tools online, e.g. BRUSELAS⁸ server (http://bio-hpc.ucam.edu/Bruselas/)  for 3D shape similarity and pharmacophore screening and Achilles for blind docking (https://bio-hpc.ucam.edu/achilles/) and how we have transferred them to the biotechnology market.

In the recent times Generative-Adversarial Networks (GANs) have captured considerable attention within the scientific community as a promising tool to discover rich causal models describing the behaviour of a variety of complex systems in physics, engineering, biology and society. In this talk, I shall describe our current attempts to put the power of GANs at the service of covid research, particularly for the detection of cryptic pockets within the covid-19 spike protein which would foster the identification of new targets for drug design. This effort is part of the CERN Against COVID-19 initiative.

Identification of druggable pockets in target proteins is the initial step for structure-based drug discovery. The most important task of a drug designer is to search for small drug-like molecules blocking these pockets on particular proteins related to some diseases. Here we present DeepSite, a machine-learning based model for pocket detection and CrypticScout, a simulation based tool for cryptic pocket detection

Large scale computing should not be achieved at the expense of computational efficiency. Computational efficiency is a measure of how well the available resources are utilized. For example, a perfectly strong scalable code can be as inefficient as one can imagine, for many reasons: a high load imbalance or a poor IPC index (Instructions per cycle). On the other hand, algorithmic efficiency is a measure of how well a physical problem is solved: a direct solver to solve a Poisson equation is not an efficient solution, but a conjugate gradient may be… In some cases, computational and algorithmic efficiencies are contradictory: this is the case of multiphysics coupling. In this context, the Jacobi and Gauss-Seidel methods are commonly used in the partitioned approach.  The Gauss-Seidel method is algorithmically more efficient than the Jacobi method, as it converges much faster. On the other hand, the Gauss-Seidel method is less computationally efficient than the Jacobi method, as the processsors of one physics are idle when the other physics is being solved.

This presentation will compare both efficiencies for the Jacobi and Gauss-Seidel methods. Finally, we will present a solution to make the Gauss-Seidel method as computationally efficient as the Jacobi method, thus making this latter one the preferred coupling method.

We present two solutions in which the open-source fluid solver Palabos is coupled to other solvers to achieve massively parallel simulations of complex problems. Both approaches lead to new open-source projects that are closely affiliated with Palabos. In the first case, Palabos is coupled to a finite-element structural solver for the fully resolved simulation of Red Blood Cells in the blood plasma. Executed on a hybrid platform, the Palabos fluid solver uses the CPUs and the npFEM structural solver the GPUs, while both solvers interact efficiently through a common MPI communication construct. In the second case, a native, GPU based fluid solver is developed with the capability to handle complex fluidic problems. The solver uses Palabos as a massively parallel Input/Output server in a client/server relationship. This time, two independent executables are created, which communicate through a trans-application MPI communicator.

Alces are collaborating with UCL Biosciences to investigate how remote, scalable HPC facilities based on public cloud could be used to provide training resources for researchers and students. In this 3-month pilot, a series of reusable, flexible compute environments will be deployed and evaluated by bioscience research groups. With the goal of developing a sustainable, automated lifecycle for deploying and leveraging appropriate teaching environments, this project aims to maximise both agility and efficiency for delivering HPC facilities exactly when and where they are needed within a strictly controlled budget.

Conception X is a venture programme and a platform that helps PhD create products and services, based on their research in deep tech. Deep tech is defined as technology based on original, cutting edge research; which is forecasted to have an enormous impact on sustainably growing the future economy. This talk will explain cover Cx is different from other programmes that commercialise research, and the rationale behind it. Then, we will explain its structure, and present a few notable case studies of successful PhD deep tech startups.

The talk will provide an overview of quantum computing and its potential applications in the pharmaceutical industry. It will provide an intro to how quantum computing methods can boost computation for complex problems in optimization, machine learning and chemistry – along with specific examples of use cases developed by Zapata.

We are a recent startup providing innovative quantum and AI solutions for molecular design. Our long term goal is to leverage quantum simulation and machine learning to design transition state based enzyme inhibitors. Here, we present our initial proof of concept and plans for future work.

Research computing in biomedical sciences presents challenges that are distinct from more traditional areas of computing. I will introduce the Biomedical Research Computing (BMRC) Facility in Oxford, which has grown organically from small beginnings with the aim of meeting the needs of biomedical science. BMRC is largely financed by coordinating the spending of grant income across multiple departments to deliver a platform that no research group could individually achieve. Rather than simply going for size, the result is a facility tightly linked to the research priorities of our supporters, and a facility that evolves as rapidly as those priorities change. BMRC has shown its flexiblity with the coronavirus pandemic: we are busier than ever with new research – and welcoming many biomedical researchers who are new to large-scale computing but nevertheless able to be productive within hours of getting accounts.