Predicting protein-protein interactions using full Bayesian network

Protein-protein interactions (PPIs) are central to the most cellular processes. Although PPIs have been generated exponentially from experimental methods ranging from high throughput protein sequences to the crystallized structures of complexes, only a fraction of interactions have been identified. It's challenging to integrate diverse datasets for computational methods. In order to predict PPIs over diverse datasets, we proposed a full Bayesian network model. First, we investigated the dihedral angle of atom C-alpha to describe flexible and rigid regions of protein structures and then design domain-domain interaction (DDI) template library to predict DDIs by the dihedral angle of atom C-alpha. Hence, both of them are viewed as the features of a full Bayesian Network (BN). Second, we used two encoding methods on sequences. The two encoding sequences can reflect both biological and physiochemical properties of proteins. Third, we also viewed gene co-expression as a feature of the BN model. Finally, we used receiver operating characteristic (ROC) to assess the performance compared to the Support Vector Machine (SVM) model.