Abstract
<jats:p>Virtual screening is a central technique in drug discovery for finding potential ligands from large chemical libraries. Conventional docking-based screening relies on a single static protein structure and therefore fails to account for the intrinsic structural flexibility of proteins, which often limits the accuracy of ligand affinity prediction. In this study, we developed a molecular dynamics (MD)-based virtual screening protocol that incorporates structural ensembles derived from ligand-bound (Holo) MD simulations to improve docking-based affinity evaluation. Using 13 MEK1 Type III inhibitors as a model system, we examined the relationship between AutoDock Vina docking scores and experimental pIC50 values. It is demonstrated that docking using a single crystal structure shows little correlation with experimental binding affinities. In contrast, ensemble-averaged docking scores calculated from representative conformations obtained by backbone and Cβ–based structural clustering of Holo MD trajectories markedly improved predictive accuracy by generating structurally meaningful receptor ensembles that capture functionally relevant variations in the ligand-binding pocket, yielding a strong correlation with experimental pIC50 values (Pearson correlation coefficient r = −0.79). Further atom contact analyses revealed that ligand scaffolds stabilize the druggable binding pocket, whereas variable substituents induce local structural fluctuations that contribute to accurate affinity prediction. The robustness of the proposed protocol was further demonstrated by its application to a larger dataset of EGFR kinase inhibitors. These results demonstrate that incorporating Holo MD–derived structural ensemble provides a practical and effective strategy for improving the accuracy and hit rates of docking-based virtual screening.</jats:p>