Comparison of Prediction Methods for Protein-Protein Interactions Using Co-Evolutionary Information

The development of computational prediction method for protein-protein interaction (PPI) using the genomic information is an important issue in bioinformatics. Mirror tree was recently proposed as a method to predict PPIs from the similarity of the phylogenetic tree or distance matrices [1]. In this method, the intensity of the co-evolution between a pair of proteins is evaluated by Pearson’s correlation coefficient between a pair of distance matrices of the proteins. However, it has been recognized that predictions by the mirror tree method include many false positives. To solve this problem, we have developed the two different methods to improve the mirror tree prediction by using partial correlation coefficients and projection operators. In our previous work, we improved the mirror tree method by excluding the information about the phylogenetic relationships from the co-evolutionary information, using the projection operator [3]. We also showed that the partial correlation coefficient is a useful statistical measure for predicting PPIs with high accuracy [2]. In this paper, we compared the prediction accuracy among the three methods, original mirror tree method, the method with projection operators and that with partial correlation coefficients. The ability and characteristics of these methods to predict PPIs were demonstrated with the dataset of pairs of physically contacting proteins.