Jacob, E. et al. eLife 10.7554/eLife.08932 (15 September 2015).

Correlated mutation analysis is an increasingly powerful approach used to help predict protein and protein-complex structures. The premise behind such methods is that mutations that occur at a given position in a protein are compensated by other mutations of residues close in space; such coevolving residues can be identified by multiple sequence alignments. However, the approach is subject to false positives stemming from indirect interactions or common ancestry. Jacob et al. report a clever way to help reduce such false positives by also considering codon-level multiple sequence alignments. They show that if the correlation is strong at the amino-acid level but weak at the codon level, it is more likely to reflect selection for a true, direct interaction, and not an indirect interaction or common ancestry. The approach can be implemented in a variety of tools to enhance their performance.