An Improved Random Forest Algorithm for Prediction of Protein-Protein Interaction

Protein-protein interaction (PPI) is a combining two or more protein because of biochemical events in any living cell. Protein domains are functional and/or structure units in a protein and consequently they are responsible for protein-protein interaction. Many machine-learning approaches with domain-based models for protein interaction prediction and their feasibility are showed. In this study, we developed a domain-based predictor based on Random Forest (RF) algorithm with the CPRS method, it is based on cost proportional roulette sampling technique and create training sample in constructing ―forest‖. Moreover, the paper applied the theory ―A protein pair is interaction with each other when they are the same function and position‖. Experimental results on Saccharomyces cerevisiae dataset show that our protein–protein interactions predictor has higher than some model with sensitivity (81.7%) and specificity (73.6%). Keywords— Random Forest, Protein-protein interaction, Domain, Roulette Sampling Technique