Secondary Metabolites Predict Diazotrophic Cyanobacteria: A Model-Based Cheminformatic Approach

Background: Nitrogen fixation (diazotrophy) is a desirable trait present in some cyanobacteria with potential applications in sustainable agriculture and chemical feedstock production. This study discovers a predictive relationship modeled between secondary metabolites and diazotrophic cyanobacteria by leveraging chemical structure similarity to identify diazotrophic strains. Methods: An algorithm was developed using chemical fingerprint similarity of metabolites curated from CyanoMetDB and evaluated with leave-one-out cross-validation on 133 manually labeled metabolites. Results: The model demonstrated strong predictive performance, achieving 88% accuracy and a ROC-AUC of 0.96. We then applied this approach to prioritize likely diazotrophic strains among 1980 unlabeled metabolites and their associated organisms, providing a rank order of most likely undetected diazotrophic strains. Toxicity analysis showed that diazotrophic-associated metabolites show similar toxicity to non-diazotrophic metabolites in rats, with less toxicity in Daphnia magna, suggesting that these metabolites are not playing a defensive role. However, these metabolites did have relatively high nitrogen presence, and many were cyclic peptides, potentially serving as signaling molecules. Conclusions: This study underscores the potential of secondary metabolites in identifying diazotrophs, even when they may not be actively demonstrating diazotrophic physiology. Discovering more diazotrophic cyanobacteria has strong implications for advancing agricultural biotechnology towards the goal of self-fertilizing crops.