Fig. 3

Diagnostic performance of the four prediction tools across antituberculosis drugs. Paired violin plots displaying sensitivity and specificity to predict drug resistance for A GenTB-Random Forest, B) GenTB-Wide and Deep Neural Network, C TB-Profiler, and D Mykrobe. E Violinplot of diagnostic performance to predict rifampicin resistance comparing isolates passing depth filters (in black) to isolates that failed the depth-filters (in gray) arranged by prediction tool. F Violinplot of diagnostic performance to predict isoniazid resistance comparing isolates passing depth filters (in black) to isolates that failed the depth-filters (in gray) arranged by prediction tool. AMK = amikacin, CAP = capreomycin, EMB = ethambutol, ETH = ethionamide, INH = isoniazid, KAN = kanamycin, OFL = ofloxacin, PZA = pyrazinamide, RIF = rifampicin, STR = streptomycin