Probe-Atom Distributions Obtained from Mixed-Solvent Molecular Dynamics Improve the Scoring of Docking Calculations

Structure-based virtual screening (VS) is widely used for the computational selection of drug candidates from compound libraries. Protein-ligand docking calculations are often performed as key steps in the early stages of this process. However, current docking calculations have limited accuracy. Thus, improvements are needed to more efficiently identify promising drug candidates. In this study, we performed mixed-solvent molecular dynamics (MSMD) simulations using four types of probe molecules to improve the accuracy of large-scale VS. We proposed a method for the modification of the docking scoring function for five selected atom classifications (XS_types). This approach integrated the grid free energy derived from the relevant atoms across the probe molecules. VS experiments conducted on nine target proteins showed improved accuracy, with the average EF1% increasing from 6.65 to 7.36. Our method may facilitate drug discovery with higher accuracy than that of conventional methods.