A Machine Learning Approach Reveals CRISPR-Cas I-F as a Genomic Marker of Antibiotic Susceptibility in Uropathogenic E. coli

Antimicrobial resistance (AMR) in Escherichia coli is a critical global health challenge, particularly in urinary tract infections, where first-line treatments are increasingly compromised. While horizontal gene transfer (HGT) via mobile genetic elements is a major driver of AMR, the genomic factors that may constrain resistance gene acquisition remain underexplored. CRISPR-Cas systems, which provide adaptive immunity against foreign DNA, could influence AMR dynamics, but their role in E. coli remains incompletely understood. We conducted a comprehensive whole-genome analysis of uropathogenic E. coli isolates, including a newly sequenced collection from Australian clinical samples and an independent, globally sourced validation cohort. Antimicrobial susceptibility profiles were integrated with CRISPR-Cas subtype classification, resistance gene burden, and mobile element content. Elastic net regression, adaptive lasso, and tree-based machine learning models were used to identify genomic predictors of resistance, with performance validated across both datasets. CRISPR-Cas subtype I-F was consistently associated with susceptibility to antibiotics commonly acquired through HGT, including trimethoprim and ampicillin, and linked to lower ARG and MGE burden. In contrast, Type I-E arrays, especially when co-occurring with orphan I-F arrays, were associated with increased resistance. These associations remained robust after adjusting for phylogroup, plasmid content, and genomic background, and were validated across datasets. Subtype-specific CRISPR-Cas systems shape antibiotic resistance profiles in E. coli, with Type I-F functioning as a potential genomic barrier to ARG acquisition. These findings highlight CRISPR array typing as a novel biomarker for AMR risk prediction and surveillance, and suggest new opportunities for leveraging CRISPR-based mechanisms to limit resistance propagation in clinical contexts.