Integrative Modular, Network-Based, and Machine Learning Framework for Predicting Accessory Genome Functions and Virulence in Escherichia coli O157:H7.

The pathogenicity of Escherichia coli O157:H7 is shaped not only by chromosomal toxins such as Stx and eae but also by virulence and resistance genes carried on plasmids. To explore the modular structure and predictive potential of these accessory elements, presence/absence data from 77 strains (70 accessory features) were analyzed. Methods included clustering using Euclidean and Jaccard distances, gene-to-gene network construction with community detection, Fisher's exact tests for associations between plasmid and virulence or antimicrobial resistance genes (AMR), random forest modeling to predict virulence labels and toxin presence (excluding direct toxin markers), and PCA for visualization. Both clustering approaches revealed broad groupings, though Jaccard clustering better captured co-occurring gene patterns. The co-occurrence network identified 12 modules, including a prominent plasmid-virulence module centered on IncF replicons, stx2, ehxA, toxB, and espP. Fisher's tests showed significant associations, notably between IncFIA and stx2c (p = 5.3 × 10-4). The Random Forest classifier achieved a cross-validated AUC of 0.853 ± 0.067, with gad, espF, IncFIA, and ehxA as key predictors. PCA explained 31.1%, 12.9%, and 10.0% of the variance across the first three components, separating plasmid-virulence module carriers from others. These findings indicate that a modular accessory genome structure contributes to the diversity of O157:H7. IncF plasmids and associated effectors form a highly interconnected subnetwork, and accessory markers independent of direct toxin genes can effectively predict virulence status, offering potential for rapid genomic surveillance of Shiga toxin-producing E. coli (STEC).