Genomic-island cassette architecture provides interpretable signal for exploratory classification of...

Enterococcus cecorum is an emerging poultry pathogen whose antimicrobial resistance and host-associated traits are often carried on genomic islands. Standard comparative genomics workflows usually reduce genomes to unordered gene inventories and may miss informative neighborhood structure within island-associated modules.

We tested whether GI (genomic island)-anchored cassette organization provides signal for distinguishing pathogenic from commensal poultry-associated E. cecorum lineages. We encoded genomic-island-anchored cassette organization as 84 genome-level summary features and evaluated this representation in 145 genomes (95 commensal, 50 pathogenic) using locked 5-fold genome-grouped cross-validation.

The cassette-summary Random Forest model achieved an area under the receiver operating characteristic curve (AUROC) of 0.918 ± 0.067, outperforming GI burden (AUROC 0.791 ± 0.050) and assembly-quality (AUROC 0.743 ± 0.015) baselines and performing similarly to a corrected AMR gene-content baseline (AUROC 0.906 ± 0.044). A conservative GI-restricted gene product presence/absence proxy achieved AUROC 0.887 ± 0.083, while a full joint-run pangenome GPA baseline remains a necessary future benchmark. Fragmentation-controlled analyses confirmed cassette signal remained informative after quality filtering (AUROC 0.827 in assemblies with ≤50 contigs; n = 91), while leave-one-BioProject-out validation yielded AUROC 0.694, indicating that deployment in novel surveillance contexts requires prospective validation. SHapley Additive exPlanations (SHAP) analysis localized discriminant signal to GI-anchored modules enriched for AMR cargo, mobility load, and GI AMR density.

These results suggest that cassette architecture captures signal consistent with biologically meaningful genomic organization beyond bulk island burden and supports its use as an interpretable exploratory representation for surveillance-oriented analysis of poultry-associated E. cecorum, while prospective validation in independent surveillance collections and a full joint-run pangenome gene presence/absence benchmark remain necessary before operational deployment claims can be made.