Structure-guided prioritization of divergent virulence and resistance candidates in an open pangenome clinical Escherichia coli isolate

Determining the biological function of uncharacterized hypothetical proteins is a key challenge in genomics. The accessory genome of pathogens houses sequence-divergent antimicrobial resistance (AMR) and virulence genes escaping standard nnotation. We present an integrated, structure-guided pangenome-to-structure computational pipeline combining pangenomic singleton filtering, structural prediction (ESMFold/AlphaFold3), homology searches (Foldseek/TM-align), Protein Language Model embeddings (ESM-2 650M), and molecular dynamics (MD) simulations (OpenMM). Applied to the clinical multidrug-resistant (MDR) extraintestinal pathogenic Escherichia coli (ExPEC) ST354 isolate QA5221, pangenome clustering across 32 genomes isolated 142 private singletons. Filtering prioritized four candidates in resistance/virulence superfamilies: (i) GNAT_KA27 ( KNGPFPPJ_02769 ), a divergent GCN5-related N-acetyltransferase (GNAT) with strong aminoglycoside affinity ( Delta G_bind = -22.91 kcal/mol) via MM-GBSA and 127-ns MD fold stability (RMSD = 2.73 ± 0.20 A); (ii) Ehly_61 ( KNGPFPPJ_00061 ), a putative enterohemolysin showing stable transmembrane topology (5.65 ± 1.04 A RMSD) in bilayer MD; (iii) OAgP_161 ( KNGPFPPJ_03161 ), a putative O-antigen polymerase ( Wzy ) in the LPS cluster (2.12 ± 0.23 A RMSD); and (iv) OAT_371 ( KNGPFPPJ_04371 ), a plasmid-associated outer-membrane autotransporter. All candidates feature low GC content (29.31–48.53%) and cluster distinctly in ESM-2 latent space. Phylogenetics confirms they form independent, deeply branching clades. This framework enables systematic prioritization of sequence-divergent accessory genes, flagging high-priority targets for experimental characterization. (GenBank: JCACXV000000000).