The study presents a computationally efficient implementation of Direct Coupling Analysis (DCA) to predict residue contacts in proteins based on sequence information alone. DCA is a method that disentangles direct and indirect correlations among amino acid compositions, allowing for the inference of spatial contacts within protein structures. The authors developed a new algorithm, mfDCA, which is significantly faster than the previous message-passing approach (mpDCA) and can analyze large protein sequences. Using this algorithm, they evaluated the accuracy of contact predictions for 131 protein domain families, finding that DCA yields a large number of correctly predicted contacts, recapitulating the global structure of the contact map for most of the domains. The predicted contacts also capture signals beyond intra-domain residue contacts, such as inter-domain interactions, alternative protein conformations, and ligand-mediated residue couplings. The results suggest that DCA can be used to facilitate computational predictions of alternative protein conformations, protein complex formation, and even de novo prediction of protein domain structures, provided a large number of homologous sequences are available. The study highlights the potential of DCA for sequence-based protein structure prediction, particularly for bacterial proteins, and opens new avenues for research in protein structure and function.The study presents a computationally efficient implementation of Direct Coupling Analysis (DCA) to predict residue contacts in proteins based on sequence information alone. DCA is a method that disentangles direct and indirect correlations among amino acid compositions, allowing for the inference of spatial contacts within protein structures. The authors developed a new algorithm, mfDCA, which is significantly faster than the previous message-passing approach (mpDCA) and can analyze large protein sequences. Using this algorithm, they evaluated the accuracy of contact predictions for 131 protein domain families, finding that DCA yields a large number of correctly predicted contacts, recapitulating the global structure of the contact map for most of the domains. The predicted contacts also capture signals beyond intra-domain residue contacts, such as inter-domain interactions, alternative protein conformations, and ligand-mediated residue couplings. The results suggest that DCA can be used to facilitate computational predictions of alternative protein conformations, protein complex formation, and even de novo prediction of protein domain structures, provided a large number of homologous sequences are available. The study highlights the potential of DCA for sequence-based protein structure prediction, particularly for bacterial proteins, and opens new avenues for research in protein structure and function.