Figure 3.

Overall accuracy (F1 score) produced by different machine learning algorithms in predicting resistance to Meropenem for (a) K. pneumoniae, (b) E. coli and Shigella, and (c) Enterobacter. “All Set” denotes all genes of a genome of interest, “Intersection Set” contains genes that consistently ranked high in their importance in discriminating resistant from susceptible strains in each round of cross-validation, and “Random Set” contains randomly sampled genes from All Set (same number of genes as in Intersection Set). Performance was assessed for each of three input datasets in a 6-fold cross-validation setting.