Obtaining highly accurate, reliable, and consistent binding free energies, across a diverse range of drug targets has been difficult to achieve for computational tools.  Free energy perturbation (FEP) is one of the few tools that has been shown to achieve highly accurate binding free energies, but you have to pay a high computational cost.  In the past this price equated to acquiring 1024 CPU cores on a massive cluster and now this same compute power is achieved by 4 Nvidia Tesla GPU's running under my desk.  The high compute price for this technique limited exploration of its applicability and reliability across diverse ligand sets and protein targets.  In this work we demonstrate our FEP/REST (Replica Exchange Solute Tempering) workflow, which harnesses the power of the NVidia GPU using a custom kernel engineered to achieve unprecedented throughput.  Using our workflow engine we are now able to run more than 200 perturbations a day on our internal Tesla powered GPU cluster, allowing us to iteratively improve the method and explore new domains for FEP, antibody design, covalent ligands, macrocyles, etc. Finally, we demonstrate how this can be scaled up to 1000 perturbations a day on the Amazon EC2 cloud using the NVidia GPU Grid, which has allowed us, for the first time, to examine how reliable and consistent binding energies can be predicted across numerous protein targets.

Harnessing the power of FEP with GPU cloud computing

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.