Clustering peptide structures through identification of commonly exposed groups

A clustering analysis method using the number of commonly exposed groups identified as a clustering criterion for a group of peptide structures generated from an in vacuo molecular dynamics simulation is presented. The number of commonly exposed groups is identified as the number of atoms of the same type which appear on vertices of groups of three dimensional convex hulls computed for groups of structures sampled and collected as blocks. Blocks of structures of high structural similarity are classified as clusters if their corresponding number of commonly exposed groups identified are larger than a preset criterion. Linkages between blocks are provided with the generation of blocks consisting of overlapping structures. However, the linkage can be eliminated by employing a minimal distance criterion for each block generated. The feasibility of this proposed clustering method is tested through a comparison of results obtained from a conventional and a hierarchical clustering method. Since change in fine structural features can be detected as the change in the number of commonly exposed groups identified, we find that the method is superior to the conventional clustering one in partitioning compact and well-separated clusters.