D-アミノ酸オキシダーゼを標的とする分子表面に対する構造ベース仮想スクリーニングとドッキング後解析

A multi-step structure-based virtual screening (SBVS) campaign was conducted on a 160,000-compound library to identify novel inhibitors targeting an outer surface region of D-amino acid oxidase (DAO), in the absence of known inhibitor binding geometries. The SBVS workflow incorporated druggability-based filtration and employed three docking engines: sievgene, DOCK 6, and AutoDock Vina. After the final screening stage and selection based on consensus among top-ranked compounds, five candidates were evaluated in vitro. Among them, one compound (N-{[(3S,3aS,6aS)-5-benzylhexahydro-2H-furo[2,3-c]pyrrol-3-yl]methyl}pyridine-3-carboximidic acid) exhibited significant inhibitory activity against DAO. To further refine the docking results and identify the most stable binding pose of the compound, we addressed the post-docking challenge using a combination of molecular dynamics (MD) simulations and machine learning-based hierarchical clustering. Poses generated from the three docking engines were classified using four clustering algorithms, and representative poses for each cluster were subjected to MD and binding free energy calculations via the molecular mechanics/generalized Born surface area (MM/GBSA) method. Notably, three poses converged to a common stable state characterized by the lowest binding free energies. The identified inhibitor in this stable state occupied a region spatially distinct from the substrate-binding site, consistent with our design strategy. Furthermore, we evaluated clustering performance based on the consistency between pose classification and calculated binding free energies. Among the tested methods, nearest neighbor and group average clustering exhibited the highest consistency. These findings demonstrate the utility of an integrated virtual screening strategy, including post-docking analysis to refine binding poses, for the robust identification of surface-binding inhibitors.