AMRvalid - Population-Aware Evaluation of Antimicrobial Resistance Predictions

Evaluates antimicrobial resistance phenotype predictions made by external models and rule-based tools against phenotypic reference results, accounting for the population structure of bacterial isolates. Provides a documented data contract and input validation; classification and probability metrics and antimicrobial susceptibility testing agreement metrics (categorical agreement, very major, major and minor errors, essential agreement) per species and antimicrobial; checks for overlap between training and test isolates; performance by genomic distance to the training data; cluster bootstrap intervals and paired model comparisons; calibration; prevalence-adjusted predictive values; and reports aligned with the TRIPOD+AI checklist (Collins et al. (2024) <doi:10.1136/bmj-2023-078378>). Organism and antimicrobial identifiers use the classes of the 'AMR' package. Models are not trained inside the package.