The long and expensive trial-and-error drug discovery process resulted in a profoundly slow pace of discovery of novel biologically active compounds. Computational molecular docking-based virtual screening provides an efficient and significantly less expensive approach to identify the drug-like lead compounds, which, upon further chemical modifications, yield drugs that can be used in therapy. However, even with state-of-the-art high- performance computational clusters, the drug virtual screening process still takes months. Furthermore, the accuracy of the virtual screening highly depends on the ability of virtual screening tools to mimic flexibility of the complex when drug is binding to its target, thereby making accuracy and speed of virtual screening to strongly depend on each other. We have previously developed a rapid and flexible protein-ligand docking program MedusaDock, which lays a solid foundation for virtual screening. Using MedusaDock, we have already discovered and in a process of commercialization of compounds for cystic fibrosis and pain. Here, we propose to extend MedusaDock to support Graphics Processing Units (GPUs) acceleration in order to leverage the powerful computing performance of GPUs and thereby expedite the virtual screening process. We have initiated research on porting MedusaDock to GPUs, which offer massive serialization of processes. In our preliminary studies, we have already improved the speed by a factor 3.5x. We expect larger improvement upon completio n of the algorithms that utilize GPUs. Here, we request for supplemental funding of a GPU cluster to perform such simulations. This cluster will be shared with our colleague Dr. Ed O?Brien in the Department of Chemistry. Adding the ability to utilize GPU clusters will create new opportunities for virtual drug screening campaigns and will help us identify new small molecule regulators of protein function.
Advances in drug discovery rely on the development of novel effective computational methodologies. Using modern hardware advancement, we propose to develop an efficient and robust computational workflow for structure-based virtual screening of very large chemical libraries.