FIG 4.

The number of subalignments and corresponding to a given model AUC for each antibiotic. Genomes were clustered based on similarity, and samples belonging to the same cluster were restricted to either the testing or the training sets in each fold of the 5-fold cross-validation. A clustering threshold of 75% k-mer similarity is shown for K. pneumoniae, and a clustering threshold of 95% is shown M. tuberculosis and S. enterica. Results for additional thresholds are shown in Fig. S7 in the supplemental material.