Prediction of E.coli protein-protein interaction sites using inter-residue distances and high-quality-index features

Abstract

We propose computational method for identification of protein-protein interaction sites using sequence and structure information. The method is trained on database of interacting proteins (DIP) for E.coli. Proteins that are known to interact are first collected from experimental results. All interacting partners are mapped onto corresponding three-dimensional structures. The training dataset for support vector machine algorithm is trained using both sequence composition and structural conformations of selected structures, if and only if both partners are composing the same complex. Our computational method is able to predict interactions for E.coli with 0.93 AUC, 0.89 sensitivity and 0.98 specificity. © 2012 Springer-Verlag GmbH Berlin Heidelberg.