PTOMSM: A modified version of Topological Overlap Measure used for predicting Protein-Protein Interaction Network

Abstract

A variety of methods are developed to integrating diverse biological data to predict novel interaction relationship between proteins. However, traditional integration can only generate protein interaction pairs within existing relationships. Therefore, we propose a modified version of Topological Overlap Measure to identify not only extant direct PPIs links, but also novel protein interactions that can be indirectly inferred from various relationships between proteins. Our method is more powerful than a naïve Bayesian-network-based integration in PPI prediction, and could generate more reliable candidate PPIs. Furthermore, we examined the influence of the sizes of training and test datasets on prediction, and further demonstrated the effectiveness of PTOMSM in predicting PPI. More importantly, this method can be extended naturally to predict other types of biological networks, and may be combined with Bayesian method to further improve the prediction.