Hopf, T.A. et al. Nat. Biotechnol. 35, 128–135 (2017).
Certain sequence features can help to determine whether a given mutation in the genome will lead to a change in function and thus potentially lend evidence for pathogenicity. Most computational methods use evolutionary conservation to predict how critical a residue is for protein function, but they do not model epistatic effects that may exist between residues. Hopf et al. incorporate this consideration in their EVmutation software by assessing sequence covariation—codependencies between amino acids or nucleotides at pairs of sites across organisms. Extensive benchmarking indicates that using these pairwise constraints in a probabilistic model provides better correlations with experimental mutational analysis than do other tools that predict the effects of mutation. The approach is less powerful in regions with low diversity, but it adds an important epistatic dimension to inform mutational prediction.