A NEW METHODOLOGY FOR IDENTIFYING INTERFACE RESIDUES INVOLVED IN BINDING PROTEIN COMPLEXES

Abstract

Genome-sequencing projects with advanced technologies have rapidly increased the amount of protein sequences, and demands for identifying protein interaction sites are significantly increased due to its impact on understanding cellular process, biochemical events and drug design studies. However, the capacity of current wet laboratory techniques is not enough to handle the exponentially growing protein sequence data; therefore, sequence based predictive methods identifying protein interaction sites have drawn increasing interest. In this article, a new predictive model which can be valuable as a first approach for guiding experimental methods investigating protein-protein interactions and localizing the specific interface residues is proposed. The proposed method extracts a wide range of features from protein sequences. Random forests framework is newly redesigned to effectively utilize these features and the problems of imbalanced data classification commonly encountered in binding site predictions. The method is evaluated with 2,829 interface residues and 24,616 non-interface residues extracted from 99 polypeptide chains in the Protein Data Bank. The experimental results show that the proposed method performs significantly better than two other conventional predictive methods and can reliably predict residues involved in protein interaction sites. As blind tests, the proposed method predicts interaction sites and constructs three protein complexes: the DnaK molecular chaperone system, 1YUW and 1DKG, which provide new insight into the sequence-function relationship. Finally, the robustness of the proposed method is assessed by evaluating the performances obtained from four different ensemble methods.