Predicting Protein Functions Based on Dynamic Protein Interaction Networks

Abstract

Accurate annotation of protein functions plays a significant role in understanding life at the molecular level. With accumulation of sequenced genomes, the gap between available sequence data and their functional annotations has been widening. Many computational methods have been proposed to predict protein function from protein-protein interaction (PPI) networks. However, the precision of function prediction still needs to be improved. Taking into account the dynamic nature of PPIs, we construct a dynamic protein interactome network by integrating PPI network and gene expression data. To reduce the negative effect of false positive and false negative on the protein function prediction, we predict and generate some new protein interactions combing with proteins’ domain information and protein complex information and weight all interactions. Based on the weighted dynamic network, we propose a method for predicting protein functions, named PDN. After traversing all the different dynamic networks, a set of candidate neighbors is formed. Then functions derived from the set of candidates are scored and sorted, according to the weighted degree of candidate proteins. Experimental results on four different yeast PPI networks indicate that the accuracy of PDN is 18% higher than other competing methods.