Bioinformatics combined with machine learning unravels differences among environmental, seafood, and clinical isolates of Vibrio parahaemolyticus

Vibrio parahaemolyticus is the leading cause of illnesses and outbreaks linked to seafood consumption across the globe. Understanding how this pathogen may be adapted to persist along the farm-to-table supply chain has applications for addressing food safety. This study utilized machine learning to develop robust models classifying genomic diversity of V. parahaemolyticus that was isolated from environmental (n = 176), seafood (n = 975), and clinical (n = 865) sample origins. We constructed a pangenome of the respective genome assemblies and employed random forest algorithm to develop predictive models to identify gene clusters encoding metabolism, virulence, and antibiotic resistance that were associated with isolate source type. Comparison of genomes of all seafood-clinical isolates showed high balanced accuracy (≥0.80) and Area Under the Receiver Operating Characteristics curve (≥0.87) for all of these functional features. Major virulence factors including tdh, trh, type III secretion system-related genes, and four alpha-hemolysin genes (hlyA, hlyB, hlyC, and hlyD) were identified as important differentiating factors in our seafood-clinical virulence model, underscoring the need for further investigation. Significant patterns for AMR genes differing among seafood and clinical samples were revealed from our model and genes conferring to tetracycline, elfamycin, and multidrug (phenicol antibiotic, diaminopyrimidine antibiotic, and fluoroquinolone antibiotic) resistance were identified as the top three key variables. These findings provide crucial insights into the development of effective surveillance and management strategies to address the public health threats associated with V. parahaemolyticus.