Answers are grounded in this paper and cite their sources.
The study of protein coevolution has undergone a remarkable transformation over the past three decades, evolving from early observations of correlated mutations in proteins to sophisticated statistical frameworks that now underpin revolutionary tools like AlphaFold. In this personal perspective, I trace this intellectual journey from our 1994 development of an operational pipeline connecting correlated mutations to protein structure prediction, through two decades of incremental refinement and frustration, to the paradigm shift brought by Direct Coupling Analysis (DCA) and the subsequent deep learning revolution. I reflect on the foundational contributions that were overlooked, the persistent challenge of disentangling phylogenetic signal from functional coevolution, and the extension of coevolutionary principles to protein-protein interactions. While celebrating the spectacular success of structure prediction methods, I argue that coevolutionary analysis remains essential, not because it predicts contacts better than AlphaFold, but because it offers a unique window into the evolutionary logic encoded in sequence variation related to protein structure, folding and function. Looking forward, I identify key unanswered questions concerning the relationship between coevolution and key areas of structural bioinformatics. This account, necessarily filtered through my own eyes, aims to complement systematic reviews by capturing the intellectual excitement, setbacks, and enduring curiosity that have driven the field for thirty years.
No takes yet. Share an insight, caveat, or question.
Alfonso Valencia (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: