High-throughput binding pose refinement through induced-fit ligand docking

Virtual ligand screening has become a fundamental step in the drug candidate discovery process. As the ligand libraries grow to billions of compounds, a workflow tailored for such a data-rich regime is required. We present the development of a high-throughput ligand pose refinement workflow, using the SARS-CoV-2 main protease active site as an example. The workflow takes in coarsely-docked ligand poses and parametrizes them for atomistic simulations with a semi-flexible, truncated protein core. We evaluate the effect of different protein core assemblies, solvation models, and integration protocols on numerical stability and success rate of the refinement. Finally, we discuss the workflow in the context of a drug discovery pipeline.