High resolution modeling of SARS-CoV2 proteome using a massive GPU-accelerated adaptive sampling strategy
, Distinguished Professor, Sorbonne Université
In this session, we'll discuss the strategy for the high-resolution modeling of the SARS-CoV-2 virus components, using molecular dynamics. We'll detail the various protein targets that are currently under study, and show how the newly developed multi-GPUs version of the Tinker-HP software can offer access to simulations with unprecedented accuracy. Using adaptive sampling and new generation many-body polarizable force fields, long (µs) molecular dynamics simulations with enhanced accuracy become possible. In that framework, the Tinker-HP package can be powered by an arbitrary large number of GPUs, reducing simulation time from years to days. The presentation will detail the supercomputer HPC implementation of the approach, as well as its cloud computing counterpart. Alongside the adaptive strategy, we'll also share recent results on multi-node MPI multi-GPUs conventional molecular dynamics simulations on very large million-atom systems. Such a large assembly of data will help to better reflect the complexity of the molecular interactions present in SARS-CoV2 components, and could be critical in drug discovery.