Species-Agnostic and Salmonella-Specific Models for Antimicrobial Resistance Prediction Using FCGR and ResNet-18

Antimicrobial resistance (AMR) prediction from bacterial genomes remains a major challenge for clinical microbiology and surveillance. We developed deep learning models based on Frequency Chaos Game Representation (FCGR) and a ResNet-18 architecture to classify resistance phenotypes directly from whole-genome assemblies. Using homology-aware clustering to prevent genomic data leakage, we compared a species-specific model for Salmonella enterica with a species-agnostic model spanning S. enterica, Escherichia coli, and Staphylococcus aureus. The Salmonella enterica model achieved high predictive accuracy, particularly for cephalosporins, while performance was lower for tetracycline and ampicillin. The species-agnostic model generalized less effectively overall but improved prediction for underrepresented species such as E. coli. Benchmarking against the gene-based tool ResFinder confirmed that FCGR-based models can approach curated predictions while capturing distributed genomic features not restricted to known resistance genes. This suggests that deep learning can highlight genomic patterns with potential biological relevance beyond established markers. Overall, species-focused models give the most reliable performance, while species-agnostic approaches can boost prediction for poorly sampled pathogens and aid AMR surveillance.