Clustering versus scoring for the identification of near-native poses in protein-ligand docking

Molecular docking is a widely used tool in structure-based drug design. A number of docking algorithms have been shown to predict accuately the crystal structure (near native) orientation of a receptor-ligand complex, and evaluate a number of putative ligands for binding affinity based on this prediction using a scoring function for binding free energy approximation. Several scoring functions have been developed in the recent past to both discriminate between docked orientations of a given ligand and score such ligands in a virtual screening experiment. In this study we investigated geometric clustering and cluster size criteria as a potential tool to discriminate between near native and false positive orientations. Our results showed that for 58% of the studied complexes, the near native orientation was within 2 A of the center of most populated docked orientation cluster, thus discrimination using cluster size could be considered a fair alternative to the scoring functions used in this study, which were able to pick the near-native like orientation 50-70% of the time.