Antibiotic-Specific Genotype–Phenotype Concordance and Cross-Database Interoperability in Public Escherichia coli/Shigella WGS-AMR Metadata

Public pathogen-genomics repositories are increasingly used for antimicrobial resistance (AMR) surveillance, yet the reliability and interoperability of database-derived genotype–phenotype inference remain incompletely characterized. This study evaluated antibiotic-specific concordance between exported AMR genotype annotations and antimicrobial susceptibility testing (AST) phenotypes in the NCBI Pathogen Detection metadata for the Escherichia coli/Shigella organism group and used BV-BRC to assess complementary phenotype-related coverage and cross-resource linkage. Frozen NCBI Pathogen Detection and BV-BRC exports were analyzed retrospectively. NCBI records were filtered for assembly accessions, AMR genotype annotations, and interpretable resistant or susceptible AST results. Prespecified antibiotic-specific mapping rules were applied. Performance metrics, including sensitivity, specificity, accuracy, balanced accuracy, positive and negative predictive values and F1-score, were calculated, and their 95% confidence intervals were estimated using 10,000 isolate-level bootstrap replicates. BV-BRC was evaluated descriptively for phenotype-related coverage, evidence type, and accession overlap. Among 539,918 NCBI records, 10,201 met prespecified criteria for assembly accession, AMR genotype annotation, and interpretable resistant/susceptible AST data. These records generated 56,141 genotype–phenotype comparisons across eight priority antibiotics. Concordance was high for tetracycline (accuracy, 98.4%; F1-score, 98.1%) and ceftriaxone (accuracy, 97.9%; F1-score, 94.6%), but lower for amoxicillin–clavulanic acid (accuracy, 73.6%; F1-score, 44.5%), indicating antibiotic-specific limits of genotype-field inference. Discordance was explicitly separated into 2524 genotype-positive/phenotype-susceptible and 923 phenotype-resistant observations with no mapped genomic evidence of resistance. BV-BRC contributed 21,298 taxonomy-strict E. coli/Shigella genomes and 4745 phenotype-linked identifiers. Although 15,214 BioSample and 10,121 assembly accessions were shared across exports, no analysis-ready overlap remained between the final NCBI genotype–AST and BV-BRC phenotype-linked subsets after eligibility filtering. Public WGS-AMR databases can support large-scale surveillance-oriented concordance analyses, but performance is antibiotic specific and depends on mapping rules, phenotype definitions, evidence provenance, and accession linkage. These estimates do not constitute clinical diagnostic validation and should not replace phenotypic AST. Because the analysis was conducted at the combined organism-group level, these estimates may mask species-, pathotype-, or lineage-specific resistance dynamics.