Development of filter functions for protein-ligand docking

Current docking methods can generate bound conformations of a ligand close to the experimentally observed structure of a protein-ligand complex. However, the scoring functions used to evaluate the potential solutions are not yet reliable enough at giving the highest ranks to the best structure predictions. One approach to this problem is the use of filter functions that are applied to all docked conformations to remove structures with certain energetically unfavorable properties. We present a computationally efficient scheme for such a postprocessing of docking results. For each of the conformations generated for a given protein-ligand complex, four properties are calculated: the fraction of the ligand volume buried inside the binding pocket, the size of lipophilic cavities along the protein-ligand interface, the solvent-accessible surface (SAS) of nonpolar parts of the ligand, and the number of close contacts between nonhydrogen-bonded polar atoms of the ligand and the protein. These four terms were used to filter out the majority of the calculated solutions and to rescore the remaining ones. On a test set of 32 protein-ligand complexes, this protocol significantly improves the accuracy of the structure predictions.