Interpretable learning algorithms enable pathogenic potential assessment and virulence-associated gene discovery of...

The presence of Vibrio parahaemolyticus (Vp) at various stages of seafood production has adversely affected public health and threatened the sustainability of the industry. Driven by the advancement of next-generation-sequencing technologies and public health data sharing initiative, an increasing volume of public Vp genomic data with metadata has become available, which serve as the foundation for building learning models to accurately differentiate isolation sources and further uncover gene-level determinants of pathogenic potential. The primary goal of this study was to develop and validate machine learning (ML) and deep learning (DL) algorithms to differentiate Vp strains from different isolation sources (clinical vs. environmental isolates) using pangenome assemblies and achieved robust and precise pathogenic potential prediction of Vp. The secondary goal of this study was to obtain critical biological insights revealing pathogenic potential-associated genes contributing to the isolation source difference from these established learning models. Based on the results, the developed learning models demonstrated strong performance, achieving an AUC greater than 0.95 in distinguishing clinical and environmental isolates using pangenome signals. Besides, the gene feature weight analysis from RF revealed the importance of specific accessory genes during Vp evolution including but not limited to functional unknown cloud genes, vspR, sctC5, and tdh1, which provides biological insights as potential future research directions. These findings essentially highlight critical importance of accessory and cloud genes in differentiating clinical and environmental isolates, and provide new insights into how recently acquired genes may contribute to pathogenic evolution of Vp. Additionally, the framework demonstrated in this study provides a cost-effective intelligent strategy by leveraging large public genomic datasets to support surveillance and risk assessment of seafood-associated pathogens.