Skip to main content

Table 1 Metrics of modified four main neural network models

From: TB-DROP: deep learning-based drug resistance prediction of Mycobacterium tuberculosis utilizing whole genome mutations

Model

Drug

AUC (Var)

Sensitivity (Var)

Specificity (Var)

Precision (Var)

NPV (Var)

wdnn_modified

Ethambutol

0.93 (0.000021)

0.89 (0.000332)

0.86 (0.000133)

0.55 (0.000311)

0.98 (0.000014)

Isoniazid

0.94 (0.000020)

0.88 (0.000179)

0.92 (0.000133)

0.83 (0.000360)

0.95 (0.000029)

Pyrazinamide

0.91 (0.000025)

0.90 (0.000400)

0.83 (0.000415)

0.42 (0.000650)

0.98 (0.000009)

Rifampicin

0.95 (0.000015)

0.89 (0.000171)

0.94 (0.000146)

0.84 (0.000598)

0.96 (0.000018)

deepamr_modified

Ethambutol

0.89 (0.000068)

0.72 (0.000144)

0.92 (0.000051)

0.62 (0.000305)

0.95 (0.000004)

Isoniazid

0.87 (0.000064)

0.75 (0.000215)

0.87 (0.000071)

0.72 (0.000125)

0.89 (0.000031)

Pyrazinamide

0.89 (0.000124)

0.70 (0.000624)

0.92 (0.000028)

0.56 (0.000190)

0.96 (0.000012)

Rifampicin

0.89 (0.000043)

0.74 (0.000314)

0.90 (0.000042)

0.74 (0.000190)

0.91 (0.000034)

cnngwp_modified

Ethambutol

0.94 (0.000024)

0.88 (0.000282)

0.89 (0.000261)

0.61 (0.001026)

0.98 (0.000009)

Isoniazid

0.94 (0.000018)

0.87 (0.000079)

0.92 (0.000053)

0.83 (0.000144)

0.94 (0.000012)

Pyrazinamide

0.93 (0.000054)

0.89 (0.000345)

0.86 (0.000040)

0.47 (0.000192)

0.98 (0.000007)

Rifampicin

0.96 (0.000007)

0.90 (0.000112)

0.93 (0.000081)

0.83 (0.000285)

0.96 (0.000012)

MLP

Ethambutol

0.93 (0.000033)

0.90 (0.000729)

0.85 (0.000242)

0.52 (0.000413)

0.98 (0.000027)

Isoniazid

0.95 (0.000031)

0.88 (0.000088)

0.90 (0.000665)

0.80 (0.001448)

0.95 (0.000014)

Pyrazinamide

0.91 (0.000091)

0.91 (0.000281)

0.81 (0.000787)

0.41 (0.001293)

0.98 (0.000007)

Rifampicin

0.95 (0.000041)

0.90 (0.000276)

0.91 (0.001085)

0.78 (0.002729)

0.96 (0.000030)

  1. Abbreviations: AUC The area under the receiver operating characteristic curve, Var variance of values of ten folds cross-validation, Positive samples are drug resistant MTB; Negative samples are drug susceptible MTB; tp: true positive, tn: true negative, fp: false positive, fn: false negative, sensitivity: tp/(tp + fn), specificity: tn/(tn + fp), precision: tp/(tp + fp), NPV: negative predictive value, tn/(tn + fn)
  2. The bold values indicate the highest performance values among four models. The values presented here were average values of tenfold cross validation. The values in the parenthesis are the variance of values of tenfold cross validation
Back to article page

BMC Genomics

ISSN: 1471-2164

Contact us