Prediction of protein-protein interaction types with amino acid index distribution and pairwise kernel SVM

Protein-protein interactions (PPIs) play a key role in many cellular processes, such as the regulation of enzymes, signal transduction or mediating the adhesion of cells. Knowing the PPI types can help the biological scientists understand the molecular mechanism of the cell. Computational approaches for identifying PPI types can reduce the time-consuming and expensive of biological experimental methods. Here, we proposed a feature extraction method, named as amino acid index distribution (AAI), to predict the PPI types (high confidence, medium confidence and low confidence). To get robust results of PPI prediction, the pairwise kernel function and support vector machines (SVM) were adopted to avoid the concatenation order of two feature vectors resulting in the unstable results for predicting PPI types. The overall success rate obtained in jackknife test was 78.62%, which is 8.52% higher than that of Chou's Isort-60D PseAAC method. The results show that the current approach is very promising for predicting PPI types.