Prediction of protein-protein interaction sites using covering algorithms

The identification of protein-protein interface residues is essential for drug design, understanding cell activity of organism. In this paper, a couple of covering algorithms are presented to predict protein-protein interaction sites by using several protein features, such as sequence profile, residue entropy and so on. These features are utilized to construct covering algorithms classifiers to identify interaction sites. The Sensitivity, Specificity, Accuracy, Correlation coefficient and F1-measure are 30.82%, 42.88%, 68.26%, 15.42%, 34.23%, respectively for alternative covering algorithm and 30.08%, 53.42%, 71.45%, 23.78%, 35.9%, respectively for neighborhood covering algorithm. It is indicated that our classifiers are robust, effective and can be applied to complement experimental techniques in studying protein-protein interactions.