Performance of AMR phenotyping methods

Table S1: Performance (F1-macro, F1-positive, F1-negative, accuracy, precision, and recall) of the KMA-based ResFinder and the BLAST-based ResFinder on each species—antibiotic combination’s whole dataset (instead of iteratively on folds ).
Table S2: Performance (F1-macro, F1-positive, F1-negative, accuracy, precision, and recall) of the Aytan-Aktug model for single species and antibiotics using the default hyperparameters, in comparison with our modified version.
Table S3: Performance (F1-macro, F1-negative, F1-positive, accuracy, precision, recall, clinical_F1-negative, clinical_precision-negative, and clinical_recall-negative) of five benchmarked methods alongside the baseline method with respect to random folds, phylogeny-aware folds, and homology-aware folds.
Table S4: A. The 13 species–antibiotic combinations for which at least one multi-factor model performed best or tied first with the single-factor models (F1-macro metric). B–J. Performance (F1-macro, F1-positive, F1-negative, accuracy, precision, and recall) of the Aytan-Aktug single-species multi-antibiotic model, the control multi-species model, and the cross-species model, as well as the LOSO cross-species models for Aytan-Aktug, Kove,r and PhenotypeSeeker.