GPU compute cluster for structural biology and novel artificial intelligence approaches to study medically important viruses

The MRC - University of Glasgow Centre for Virus Research is the largest collective of researchers in the UK dedicated to the study of viruses that infect humans and at the human-animal interface. The centre has a substantial data science portfolio that ranges in scope from structural biology to molecular evolution, transmission and epidemiological studies. We are seeking funds to support the purchase of a high-performance graphics processing unit (GPU) based computer cluster for the development and implementation of artificial intelligence (AI) tools with a specific focus on virus structure and the application of structural biology tools to the study of virus diversity and evolution. Virus protein structures underlie fundamental interactions with their hosts in infection, replication, and immune evasion. Evolutionary change in these structures can therefore impact susceptibility of novel host species during spillover, virulence, and escape from drugs or vaccines. Examples include SARS-CoV-2 persistence in the human population due to ongoing antigenic change, and the risk of avian influenza adapting to humans. Machine learning and AI are transforming computational biology research. Notably, innovators in the field of protein structure prediction and design were awarded the 2024 Nobel prize in Chemistry for the impact they have had on life sciences through the development of AI approaches such as AlphaFold and RFDiffusion. Within the CVR, innovative applications of AI have recently included development of tools to identify the most likely natural host of viruses and whether they are likely to infect humans, as well as the most likely animal hosts of SARS-COV-2  based solely on viral sequence data, the use of protein language models to identify mutations that are likely to impact on viral phenotypes in human hosts and to predict virus-host protein interactions. We have also applied emerging AI protein structure prediction methods to investigate the deep evolutionary links between viruses. We have created a unique database of >85k viral protein structure predictions that spans >4.4k viruses of animals and humans (https://viro3d.cvr.gla.ac.uk/). This facilitates structure informed discovery of conserved functions across diverse virus species. In structural biology, and in particular cryogenic electron microscopy (cryo-EM) emerging AI tools for image denoising and structural motif detection are allowing researchers to begin to disentangle and interpret the information rich data produced by 3D imaging of the cell, enabling us to map the distribution of proteins, visualising biologically important processes in their native environment. AI automated model-building tools and enhanced refinement techniques are enabling faster and more robust structure determination of ever-smaller complexes. To support our work on the structural biology of viruses, and better integrate structure into computational models of virus evolution and host range, we aim to install a GPU cluster equipped with two types of GPU nodes, one optimised for training large machine learning models capable of representing viral genome sequences in new and richly-parameterised spaces, the other equipped for rapidly computing experimental and predicted protein structures based on established or newly developed models. Together this resource will ensure CVR researchers remain competitive in these fast-moving and exciting fields, supporting the development and implementation of new tools to underpin our pandemic preparedness mission.