Correction for phylogeny, small number of observations and data redundancy improves the identification of coevolving amino acid pairs using mutual information.

CRISTINA MARINO BUSLJE; JAVIER SANTOS; JOSE MARIA DELFINO; MORTEN NIELSEN

BIOINFORMATICS (OXFORD, ENGLAND)

Oxford Journals

Año: 2009 vol. 25 p. 1125 - 1131

1367-4803

ABSTRACTMotivation: Mutual information (MI) theory is often applied to predictpositional correlations in a multiple sequence alignment (MSA) tomake possible the analysis of those positions structurally or functionallyimportant in a given fold or protein family. Accurate identificationof coevolving positions in protein sequences is difficult due tothe high background signal imposed by phylogeny and noise. Severalmethods have been proposed using MI to identify coevolvingamino acids in protein families.Results: After evaluating two current methods, we demonstrate howthe use of sequence-weighting techniques to reduce sequence redundancyand low-count corrections to account for small number ofobservations in limited size sequence families, can significantly improvethe predictability of MI. The evaluation is made on large setsof both in silico-generated alignments as well as on biological sequencedata. The methods included in the analysis are the APC(average product correction) and RCW (row-column weighting)methods. The best performing method was APC including sequence-weighting and low-count corrections. The use of sequencepermutations to calculate a MI rescaling is shown to significantlyimprove the prediction accuracy and allows for direct comparison ofinformation values across protein families. Finally, we demonstratehow a lower bound of 400 unique sequences is needed in an MSAin order to achieve meaningful predictive performances. With ourcontribution, we achieve a noteworthy improvement on the currentprocedures to determine coevolution and residue contacts, and webelieve that this will have potential impacts on the understanding ofprotein structure, function and folding.