Toward the fast blind docking of a peptide to a target protein by using a four-body statistical pseudo-potential

In vitro molecular evolution creates a lot of peptide aptamers that bind to each target protein. In many cases, their binding sites on a protein surface are not known. Then, predicting the binding sites through computation within a reasonable time is desirable. With this aim, we have developed a novel system of fast and robust blind docking of a peptide to a fixed protein structure at low computational costs. Our algorithm is based on the following scheme. Representing each of the amino acid residues by a single point corresponding to its side-chain center, the structure of a target protein and that of a ligand peptide are coarse-grained. The peptide, which is described as a flexible bead model, is movable along the grid-points which are set surrounding the protein. An arbitrary state of the protein-peptide complex is subjected to Delaunay tessellation. Then, the fitness of a peptide-coordination to the protein is measured by a four-body statistical pseudo-potential. Through 1000 trials of a simple hill-climbing optimization, the best 15 peptide-coordinations with the 1st-15th highest fitness values are selected as candidates for the putative coordination. Retrieving the available 28 protein-peptide complexes from the Protein Databank, we carried out the blind docking test for each system. The best 15 peptide-coordinations fell into several clusters by the cluster analysis based on their spatial distribution. We found that, in most cases, the largest cluster or second largest cluster correspond to nearly correct binding sites, and that the mean (+/- standard deviation) of GTGD over all the 28 cases is 4.8 A(+/-3.8 A), where GTGD represents the distance from the putative binding site to the correct binding site.