Abstract—Proteins fulfill their functions by binding with molecular compounds called ligands. This research automatically extracts a binding site from the surface of a protein. A binding site candidate can be extracted as the local portion that satisfies the following two requirements. One is the structural similarity among proteins that bind the same kind of ligands. The other is the structural dissimilarity between a binding site in a protein and any local surfaces in the proteins that bind to any other ligands. By representing a protein molecular surface as a graph, the binding site extraction problem can be regarded as an optimal subgraph detection problem in which the best subgraph is extracted that satisfies the above requirements. However, if two ligands are different but have a partly similar structure, the binding sites of the proteins that bind these ligands often resemble each other. In such situations, an optimal graph may not present the binding site. Therefore, we introduce the concept of group integration, in which more than one group with similar ligands partners is regarded as a positive group. As a result of group integration, the number of proteins in a positive group and in a negative group is changed. Therefore, based on the distance from the virtual worst subgraph, an evaluation function is introduced to compare subgraphs with and without group integration. We clarified the effectiveness of binding site extraction with group integration through an experiment with 37 proteins.

Index Terms—Binding site extraction, graph mining, group integration, protein surface.

The authors are with the Graduate School of System Informatics, Kobe University, Japan (e-mail: mitsui@cs25.scitec.kobe-u.ac.jp, ohkawa@kobe-u.ac.jp).

 

Cite: Takuma Mitsui and Takenao Ohkawa, "Binding Site Extraction by Detecting Optimal Graphs from Protein Molecular Surfaces," International Journal of Bioscience, Biochemistry and Bioinformatics vol. 4, no. 1, pp. 28-32, 2014.