Improving Binding Free Energy Predictions with Swap Monte Carlo for Water Sampling

Abstract Water molecules located within and surrounding the binding cavity can substantially affect ligand binding affinity. Proper sampling of these water molecules ensures that their contributions to the free energy landscape are accurately accounted for. In the pursuit of more accurate binding free energy calculations, we have developed a novel Swap Monte Carlo (SwapMC) method specifically designed for cavity water sampling. The SwapMC method aims to enhance the sampling efficiency by facilitating the movement of water molecules in and out of the protein cavity, thereby ensuring a comprehensive exploration of water distributions. By leveraging GPU power to perform Monte Carlo moves for water molecules in parallel across multiple sites, and integrating SwapMC with NPT simulations within the Uni-FEP framework, we have observed significant improvements in the accuracy of relative binding free energy calculations, all while maintaining computational efficiency. Our results demonstrate that SwapMC achieves performance comparable to Grand Canonical Monte Carlo (GCMC) methods in water-related test cases, offering a robust and efficient alternative for addressing the challenges associated with cavity water sampling in molecular dynamics.