Abstract:Automated annotation of protein function is challenging. As the number of sequenced genomes rapidly grows, the overwhelming majority of proteins can only be annotated computationally. Under new conditions or stimuli, not only the number and location of proteins would be changed, but also their interactions. This dynamic feature of protein interactions, however, was not considered in the existing function prediction algorithms. Taking the dynamic nature of protein interactions into consideration, we construct a dynamic weighted interactome network (DWIN) by integrating protein-protein interaction (PPI) network and time course gene expression data, as well as proteins' domain information and protein complex information. Then, we propose a new prediction approach that predicts protein functions from the constructed dynamic weighted interactome network. For an unknown protein, the proposed method visits dynamic networks at different time points and scores functions derived from all neighbors. Finally, the method selects top N functions from these ranked candidate functions to annotate the testing protein. Experiments on PPI datasets were conducted to evaluate the effectiveness of the proposed approach in predicting unknown protein functions. The evaluation results demonstrated that the proposed method outperforms other competing methods.

A New Method for Predicting Protein Functions from Dynamic Weighted Interactome Networks