Abstract
<jats:p>Microbial communities are shaped by secreted metabolites that mediate ecological interactions, yet predicting these interactions from genomic sequences remains difficult, because the specific recognition between co-functional metabolites (CFMs), such as siderophores, and their receptor proteins (Rec) cannot be inferred from gene annotation alone. This difficulty arises from three factors: the prevalence of Rec-mediated exploitation, the lack of high-accuracy functional annotations, and the absence of genomic co-localization between functionally paired CFM-Rec in Gram-positive bacteria. Here we present the Coevolution-based Interaction Model (CIM), an automated framework that maps specific CFM-Rec pairings directly from uncurated genomic datasets. Using a dynamic joint optimization strategy that accounts for exploitation asymmetry and avoids combinatorial explosion, CIM identifies functional pairings solely through evolutionary covariation. We validated this approach by reconstructing macroscale iron scavenging networks across nine bacterial taxa. Experiments confirmed that CIM can bridge genomic distances exceeding 3 Mb in the Gram-positive genus Rhodococcus to identify unlinked cognate receptors, and can accurately predict cross-utilization by exploiter strains despite substantial receptor sequence heterogeneity in Burkholderiaceae and Rhizobiaceae. Finally, topological analysis of the reconstructed networks shows that siderophore exploitation acts as a universal topological glue, fusing fragmented microbial populations into highly connected communities, and that the exploitability of siderophore production reverses depending on network modularity. CIM thus offers a scalable, sequence-to-ecology approach for predicting interactions mediated by secondary metabolites in microbial communities.</jats:p>