Predicting co-evolving pairs in Pfam using information theory where entropy is determined by phylogenetic mutations

The accurate prediction of co-evolving pairs in protein sequences plays an important role in tertiary protein structure prediction and protein engineering. Using information theory to detect coevolving pairs is impacted by the phylogenetic effect on entropy measurements. Mutual Information (MI) is used to detect co-evolving pairs in a protein family by re-sampling based on mutation events in the phylogenetic tree (RPE). The predictive quality of co-evolving pairs with high mutual information (Z>=4), a sequence distance > 10 and within 12 angstroms using the RPE method is on average 81% in a Pfam family. The accuracy of detecting co-evolving pairs without RPE is 56%. This study represents the first known analysis of mutual information to detect co-evolving pairs in the full Pfam data set. Results for each protein family in Pfam and corresponding ribbon model graphing the MI relationships can be found at http://www.proteinx3d.com.