Prediction of protein-protein interactions using subcellular and functional localizations

Abstract

Protein-protein interaction (PPI) plays an important role in the living organisms, and a major goal of proteomics is to determine the PPI networks for the whole organisms. So both experimental and computational approaches to predict PPIs are urgently needed in the field of proteomics. In this paper, four distinct protein encoding methods are proposed, based on the biological significance extracted from the categories of protein subcellular and functional localizations. And then, some classifiers are tested to prediction PPIs. To show the robustness of classification and ensure the reliability of results, each classifier is examined by many independent random experiments of 10-fold cross validations. The model of random forest achieves some promising performance of PPIs. © 2010 Springer-Verlag.