Predicting Binding Poses and Affinities in the CSAR 2013-2014 Docking Exercises Using the Knowledge-Based Convex-PL Potential.

The 2013-2014 CSAR docking exercise was the opportunity to assess the performance of the novel knowledge-based potential we are developing, named Convex-PL. The data used to derive the potential consists only of structural information from protein-ligand interfaces found in the PDBBind database. As expected, our potential proved to be very efficient in the near-native pose detection exercises, where we correctly predicted two near-native poses in the 2013 exercise and also ranked 22 near-native poses first and 2 second in the 2014 exercise. Somewhat more surprisingly, we obtained a fair performance in some of the CSAR affinity ranking exercises, where the Spearman correlation coefficients between our predictions and the experiments are greater than 0.5 for several protein-ligand sets. Nonetheless, affinity prediction exercises turned out to be a challenge, and significant progress in the development of our method is needed before we can successfully predict binding constants.