Prediction of protein-protein interactions in yeast using SVMs with genomics/proteomics information and feature selection

In current Proteomics research, prediction of protein-protein interactions (PPIs) is one of the main goals, since PPIs explain most of the cellular biological processes. In the present work, we propose a method for prediction of protein-protein interactions in yeast. Our proposal is based on the well-known classification paradigm called support vector machines and a well-known feature selection method (Relief) using genomics/proteomics information. In order to obtain higher values of specificity and sensitivity in predicting PPIs, we use a high reliable set of positive and negative examples from which to extract a set of proteomic/genomic features. We also introduce a similarity measure for pairs of proteins to calculate additional features from well-known databases, that allow us to improve the prediction capability of our approach. After applying a feature selection method, we construct SVM classifiers that obtain a low error rate in the prediction for each pair of proteins. Finally, we analyse and compare the prediction quality of the method proposed with other high-confidence datasets from other works.