CAREER: Origins of microbial niche adaptation

Microbes are integral to nearly every ecosystem and every human body, and different microbes have adapted to thrive in radically different niches: those on human skin must survive host antimicrobial defenses, while those in mine runoff must tolerate acidic conditions and heavy metals. Understanding how microbes adapt and thrive in each hostile niche could help harness their tools for societal benefit. For example, identifying the genes underlying niche adaptation would help researchers understand how members of the human microbiota contribute to health or disease, how to safely engineer microbes to convert waste into useful products, and how to rapidly forecast the capabilities or risks of a newly sequenced microbe. This project develops artificial intelligence and machine learning methods, including protein language models, that detect complex genomic patterns linked to niche adaptation across bacterial lineages such as Staphylococcus, Pseudomonas, and Mycobacterium. As these tools are built and released, the project will train the next generation of computational biologists through R-Ladies+ community programs and an immediate, hands-on workshop series modeled on the Carpentries training framework, broadening participation in data-intensive biology. Together, this work will help the field of microbiology rapidly understand how microbes adapt to new environments, strengthening national health, biotechnology innovation, and U.S. training and leadership in AI/ML research. This project addresses the microbial genotype-to-phenotype gap, with an emphasis on the evolutionary strategies underlying niche adaptation, such as host switching, saltwater adaptation, and pathogen vs. commensal adaptation. Whole-genome sequencing datasets have grown exponentially, enabling analysis of billions of nucleotides across tens of thousands of genomes per species, but this scale has created new challenges in detecting meaningful signal. A foundational data harmonization effort will unify fragmented public genomic and phenotypic data, including natural language processing methods to clean and impute messy metadata, into a FAIR-compliant knowledgebase supporting all downstream analyses. Research Theme 1 develops and compares classical machine learning and protein language model approaches to identify interpretable, ranked gene, protein, and functional signatures of host adaptation, seeking both species-specific and broadly conserved motifs across host niches. Research Theme 2 constructs multipartite network models using link prediction and graph neural networks to integrate ecophysiological traits with genomic features and identify subtle predictors of niche specialization. Together, these themes will produce a systematic, evolutionarily grounded understanding of how bacteria colonize, adapt to, and thrive in new environments, yielding testable predictions for lab validation and open-source tools for the broader research community. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.