Prediction of Protein-Protein Interaction using validated domain-domain interaction

Abstract

Protein-Protein Interaction (PPI) is of biological interest since it is involved in a number of cellular processes such as metabolic pathways, immunological recognition. The knowledge of PPI is important for the investigation of intracellular signaling pathways, modeling of protein complex structures and for gaining overview of various biochemical processes. PPI prediction identifies and catalogs physical interactions between pairs or group of proteins. Recently, various methods of predicting PPI using domain information are proposed. Here a two-class support vector machine based method PPI_DOMAIN is presented exploiting interaction between constituent domains in protein pairs. Unlike the most existing methods which consider only single domain protein pairs, this method is capable of exploring multi-domain proteins where all possible combination of constituent domain pairs is considered. This is done by validating the domain pairs from DOMINE database and make predictions based on domain-domain interaction. PPI_DOMAIN is designed with two class support vector machine (SVM) using domain information with different kernels (Linear, Polynomial and Radial basis function). Interacting protein pairs are taken from Database of Interacting Protein (DIP). Using four-fold cross-validation this classifier achieves accuracy of 91.22% with precision/specificity of 95.76% and recall/ sensitivity of 86.43%.