Data-Driven Approaches for Molecular Docking

University Of California, San Francisco, San Francisco CA

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): Molecular docking approaches for drug discovery are used for predicting ligand binding modes, ligand binding affinities, and for computational screening to identify new specific ligands for biological targets of pharmaceutical relevance. Current approaches are characterized by high inter-target variability and sharp limitations in ability to make accurate predictions for new ligands that are structurally different from known ones. Approaches that model protein flexibility directly or approach simulation-level detail in complex molecular systems have shown some success. However, they are generally applicable only in low throughput and in cases where high-quality experimentally determined protein target structures are available. This is in part why "me-too" drugs dominate the pharmaceutical marketplace and development pipeline. Such drugs generally bring much less pharmacological novelty to patient treatment than structurally novel therapeutics. We propose an integrated set of methods for molecular docking that treats protein flexibility in a serious manner, is computationally efficient enough for wide use, and which offers the opportunity to effectively use docking in cases where few experimental structures exist for a biological target of interest. Our recent work has established an approach to treating protein flexibility in docking that addresses large protein movements by considering multiple experimental structures and small ligand-dependent movements by protein/ligand complex relaxation beginning from many putative ligand dockings. We have also established an approach for de novo protein pocket induction that constructs a binding site based solely on ligand binding data that is capable of making accurate predictions of binding geometry and binding affinity for structurally novel ligands. Our proposed work will combine these approaches. In cases where protein structural information is available, experimentally determined structures will undergo additional sampling, followed by refinement of the binding pockets based on ligand binding data in order to improve the predictions obtained from docking based upon our existing methods developed for de novo pocket construction. In addition to data-driven pocket refinement, the proposed effort requires improvement in our scoring functions for docking, taking into account vastly more data and explicit modeling of protein flexibility and of the unbound states of proteins and ligands. We will also put significant effort into algorithmic improvements that will result in typical run-times on common single-processor hardware of several minutes per ligand to yield predictions of binding geometry and affinity. We believe that a widely applicable, genuinely predictive, and computationally tractable modeling approach to docking will substantially improve drug discovery in practice. These methods will facilitate identification of novel lead compounds from directed lead optimization and computational screening exercises.