Effectiveness of correlation analysis in identifying protein residues undergoing correlated evolution.

Various methods for detecting correlation between sites were evaluated by ascertaining their ability to discriminate positively correlated sites from background correlation at randomly evolved sites. A model for generating pairwise correlations of different degrees is also described. An assortment of physicochemical vectors and similarity and difference matrices were used to discriminate correlated change. There was little difference in effectiveness between the different matrices, but there were significant differences between the matrices and the physicochemical vectors. It is shown that all methods investigated exhibit significant inability to screen out background correlation, particularly in the presence of phylogenetic relatedness between the sequences. Methods using the matrices are unable to distinguish positively correlated from negatively correlated, or compensatory, replacements.