Skip to main content

Table 1 Diagnostic accuracy of GenTB RandomForest and GenTB wide and deep neural network compared with two other leading prediction tools on a depth filtered dataset

From: GenTB: A user-friendly genome-based predictor for tuberculosis resistance powered by machine learning

Drug name

Phenotype

GenTB - RF

GenTB - WDNN

Mykrobe

TB-Profiler

 

R (n)

S (n)

Sensitivity (95% CI)

Specificity (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

Isoniazid

6,043

13,112

91% (91 to 92)

98% (97 to 98)

90% (89 to 91)

99% (99 to 99)

87% (86 to 88)

98% (98 to 98)

91% (90 to 92)

98% (97 to 98)

Rifampicin

5,068

14,474

93% (93 to 94)

98% (98 to 98)

88% (88 to 89)

99% (99 to 99)

90% (89 to 91)

98% (98 to 99)

92% (91 to 93)

98% (98 to 99)

Ethambutol

2,936

12,362

86% (85 to 87)

92% (92 to 93)

82% (80 to 83)

93% (93 to 94)

79% (77 to 80)

93% (93 to 94)

86% (85 to 88)

92% (92 to 93)

Pyrazinamide

508

1,544

79% (76 to 83)

94% (93 to 95)

80% (79 to 82)

95% (94 to 95)

72% (71 to 74)

98% (97 to 98)

83% (80 to 86)

96% (96 to 97)

Amikacin

618

3,458

67% (63 to 71)

99% (99 to 100)

66% (62 to 70)

99% (99 to 100)

63% (60 to 67)

99% (99 to 100)

55% (51 to 59)

99% (99 to 100)

Capreomycin

648

3,733

63% (59 to 67)

97% (97 to 98)

57% (53 to 61)

98% (98 to 99)

60% (56 to 64)

98% (98 to 99)

56% (52 to 60)

96% (95 to 96)

Ethionamide

502

1,094

67% (63 to 72)

78% (75 to 80)

-

-

-

-

70% (66 to 74)

73% (70 to 76)

Kanamycin

576

3,707

68% (64 to 72)

99% (98 to 99)

66% (62 to 70)

100% (99 to 100)

66% (63 to 70)

99 (99 to 100)

68% (64 to 71)

98% (98 to 99)

Streptomycin

2,126

4,968

82% (80 to 83)

89% (88 to 90)

87% (85 to 88)

87% (86 to 88)

68% (66 to 70)

95% (95 to 96)

71% (70 to 73)

95% (95 to 96%)

Ofloxacin

743

4,038

68% (65 to 72)

99% (98 to 99)

62% (58 to 66)

96% (95 to 96)

62% (58 to 65)

99% (98 to 99)

67% (63 to 70)

98 (98 to 99)