A multi-scale coevolutionary approach to predict interactions between protein domains

International audienceInteracting proteins and protein domains coevolve on multiple scales,from theircorrelated presence across species, to correlations inamino-acid usage. Genomic databases provide rapidly growing data forvariability in genomic protein content and in protein sequences, callingforcomputationalpredictions ofunknowninteractions.We first introduce the concept of direct phyletic couplings,based onglobal statistical modelsofphylogenetic profiles. They strongly increase the accuracy of predicting pairs of related proteindomains beyond simpler correlation-basedapproaches like phylogenetic profiling(80% vs. 30-50% positives out ofthe 1000 highest-scoringpairs). Combined with thedirect coupling analysis of inter-protein residue-residue coevolution, we provide multi-scale evidence fordirect but unknown interaction between protein families. An in-depth discussion shows these to bebiologically sensible and directly experimentallytestable. Negative phyletic couplings highlightalternative solutions for the same functionality, including documented cases of convergent evolution.Thereby our workprovesthe strong potential of global statistical modeling approaches to genome-wide coevolutionaryanalysis, far beyond the established use forindividual protein complexesand domain-domain interactions