Exploring Large Scale Receptor-Ligand Pairs in Molecular Docking Workflows in HPC Clouds

Computer-aided drug design techniques are important assets in pharmaceutical industry because of their support for research and development of new drugs. Molecular docking (MD) predicts specific compound's binding modes within the active site of target proteins. Since MD is a time-consuming process, existing approaches reduce the number of receptors or ligands in docking by evaluating only small sets of compounds. This restriction in the search space reduces the chances to uniformly cover the diverse space of compounds and misses opportunities to recognize whether new drugs can be identified. Another difficulty with large-scale is analyzing the results, e.g. browsing all directories manually to find which pairs were docked successfully. To address these issues we explored the potential of data provenance analysis and parallel processing of SciCumulus, a cloud Scientific Workflow Management System. We present SciDock, a molecular docking-based virtual screening workflow and evaluate its execution using 10,000 receptor-ligand pairs related to proteases enzymes of protozoan genomes. The overall performance of SciDock using 32 cores, in cloud virtual machines, reaches improvements up to 95.4% when running SciDock with AutoDock and 96.1% when running SciDock with Vina. We show how data provenance improved the result analysis and how it may indicate potential proteases drug targets for protozoan treatments.