Zero Hunger (SDG 2), Good Health & Well-being (SDG 3)
Enhancing structure-based virtual screening with machine learning
Francesco Gentile* Department of Chemistry and Biomolecular Sciences, University of Ottawa, Canada
Virtual make-on-demand libraries of billions of synthesizable molecules are paving the way for unprecedented opportunities in early-stage drug discovery. Computational methods, especially molecular docking, have emerged as the leading strategies for prioritizing novel ligands from these databases. Unfortunately, the computational cost associated with docking and its suboptimal performance against unconventional targets make it difficult to fully exploit the therapeutic and biological potentials of these growing databases, which consequently remain out of reach for the scientific community. Here, we summarize our recent efforts to develop computational strategies to alleviate some of these emerging challenges. We introduce Deep Docking (DD), a method that leverages active learning to efficiently predict docking scores of small molecules from their chemical structures, enabling a hundred- fold reduction of time and resources required for virtual screening. Successful applications of DD are highlighted, especially in the context of the discovery of inhibitors for the SARS-CoV-2 main protease. Recent efforts of our group to improve the success rate of virtual screening campaigns with deep neural networks are also presented. These results highlight the rising role of machine learning coupled with physics-based methods to accelerate the discovery of novel chemistry for drug targets and to democratize drug discovery.
P15
© The Author(s), 2023
Made with FlippingBook Learn more on our blog