Prediction of protein-protein interactions through support vector machines

In this paper, a SVM-based method is implemented for the prediction of protein-protein interactions. This model is initially trained with a set of over 69.000 pairs of protein sequences based on documented positive interactions. Then, a cross-validation method is performed for estimating the accuracy of the system, showing acceptable performances in terms of sensitivity, specificity and geometric mean. The results are approximately balanced and the overall performance if around 70% classified through a pairwise kernel and the parameters are set through an particle swarm optimization meta-heuristic and showing promising results for the field of bioinformatics.