Machine learning based protein-protein interaction prediction using physical-chemical representations

Many proteins can interact with other proteins to perform specific functions. Predicting those interactions is important in order to analyze signaling pathways or to define the influence of a specific protein in some diseases. This work proposes the implementation of Support Vector Machines (SVM) for the prediction of protein-protein interactions using physical-chemical features taken from AA index. This algorithm was trained with a set of over 10.000 positive interactions from DIP database, and the same number of negative interactions through random permutations. The obtained results demonstrate that these features can provide useful information for the training set in order to improve the quality of the classification. Additionally, tunning the parameters of the SVM with Particle Swarm Optimization, lead to significantly improve the performance of the machine (greater than 70%), in comparison to recent studies.