FIS-PNN: a hybrid computational method for protein-protein interactions prediction using the secondary structure information

Abstract

The study of protein-protein interactions (PPI) is an active area of research in biology because it mediates most of the biological functions in any organism. This work is inspired by the fact that proteins with similar secondary structures mostly share very similar three-dimensional structures, and consequently, very similar functions. As a result, they must interact with each other. In this study we used our approach, namely FIS-PNN, to predict the interacting proteins in yeast from the information of their secondary structures using hybrid machine learning algorithms. Two main stages of our approach are similarity score computation, and classification. The first stage is further divided into three steps: (1) Multiple-sequence alignment, (2) Secondary structure prediction, and (3) Similarity measurement. In the classification stage, several independent first order Sugeno Fuzzy Inference Systems and probabilistic neural networks are generated to model the behavior of similarity scores of all possible proteins pairs. The final results show that the multiple classifiers have significantly improved the performance of the single classifier. Our method, namely FIS-PNN, successfully predicts PPI with 96% of accuracy, a level that is significantly greater than all other sequence-based prediction methods.