Predicting Protein-Protein Interactions with Weighted PSSM Histogram and Random Forests.

The prediction of protein-protein interactions is one of the most important and challenging problems in computational biology. Because position specific scoring matrix PSSM encodes the evolutionary conservation information of a protein, the PSSM-derived features have been widely used to predict protein-protein interaction residues in previous studies. In this paper, we developed a novel method to extract feature, called weighted PSSM histogram, from the PSSM of a protein by introducing the concept of histogram in digital image processing field. Based on the extracted weighted PSSM histogram and several traditional features, we trained a random forests prediction model. Experiment results on benchmark datasets demonstrated the efficacy of the proposed method.