Predicting Co-Complexed Protein Pairs Based on Communication Model Using Diverse Biological Data

Protein-protein interactions play key role in many fundamental biological processes, and comprehensively identifying them represents a crucial step towards systematically defining their cellular roles. Machine learning techniques have been employed to predict protein-protein interactions. One of such approaches is Naive Bayes approach which assumes conditional independence between features. And such problems as suffering from the missing value problems or being prohibitively time-consuming prevent them from being applied to predict PPI as readily as NB. In this work, we frame predicting PPI as a communication problem, and we train a classifier based on channel model (CBCM) to discriminate between pairs of proteins that are co-complexed and pairs that are not. We theoretically demonstrate that NB can be unified into CBCM in certain condition and also experimentally validate that CBCM is an effective approach for predicting co-complexed protein pairs from integrating diverse biological data. Our study suggests that PPI prediction problem can be effectively solved from the point view of communication problem.