Large-Language-Model-Assisted Microbial Source Tracking Reveals Mechanisms of Tropical Cyclone Impacts on Microbial Water Quality in Coastal Environments

Extreme weather events increasingly threaten coastal water quality, yet the mechanisms by which tropical cyclones impair microbial conditions remain poorly quantified. We develop a Large Language Model–Assisted Microbial Source Tracking (LAMST) framework to trace the origins of microbial threats—fecal indicator bacteria (FIB), pathogens, and antimicrobial resistance genes (ARGs)—and apply it to coastal waters impacted by Hurricane Milton along Florida’s Gulf Coast. LAMST integrates 16S rRNA sequencing with species-level metadata from the NCBI BioSample database to probabilistically classify detected taxa as marine, terrestrial, or wastewater in origin. Across 30 sites and three time points (1 week, 2 weeks, and 7 months post-storm), twelve analytes—spanning total bacteria, FIB, pathogens, and ARGs—were quantified by digital PCR. Both terrestrial and marine bacteria showed significant increases in total bacterial concentrations in the coastal waters following the hurricane, indicating concurrent mobilization of land-derived inputs and marine sources. Site fixed-effects regressions showed that terrestrial and wastewater bacteria were strongly associated with enterococci and two ARGs (sul2 and tetA), whereas marine bacteria correlated with Vibrio parahaemolyticus. Despite rapid declines in enterococci within two weeks, several pathogens and ARGs remained elevated, underscoring the limitations of FIB alone as a regulatory indicator of post-storm microbial impairment. LAMST provides a quantitative and generalizable framework for source-resolved microbial tracking without reliance on regional reference sequences, offering actionable insights for post-storm water-quality assessment and coastal resilience planning.