Table 2.
Performance of best models restricting to only important mutations for classification.
| Drug | INH | EMB | RIF | PZA | FDR-TB | MDR-TB |
|---|---|---|---|---|---|---|
| Best model | IF3 (0.001) + MLRF | IF3 (0.005) + SLRF | IF3 (0.001) + MLRF | IF1 (0.001) + SLRF | IF3 (0.01) + SLRF | IF3 (0.001) + MLRF |
| Number of mutations | 37 | 17 | 37 | 32 | 16 | 37 |
| Sensitivity | 92.88 (↓0.28) ± 0.93 | 91.10 (↓0.65) ± 1.76 | 92.19 (↓0.07) ± 1.10 | 84.73 (↓2.54) ± 2.49 | 91.74 (↑4.16) ± 3.37 | 93.76 (↑0.06) ± 1.33 |
| Specificity | 97.88 (↑0.09) ± 0.31 | 92.70 (↑1.12) ± 0.51 | 97.77 (↓0.22) ± 0.52 | 92.83 (↓2.12) ± 0.52 | 90.06 (↓2.92) ± 0.61 | 97.38 (↓0.07) ± 0.49 |
| AUC | 95.48 (↓0.53) ± 0.40 | 91.90 (↑0.20) ± 0.82 | 94.98 (↓1.02) ± 0.53 | 88.78 (↓0.21) ± 1.17 | 90.90 (↑0.62) ± 1.56 | 95.47 (↓0.11) ± 0.62 |
The number of mutations used for the classification, best model and performance for INH, EMB, RIF, PZA, MDR-TB, and FDR-TB are shown. Increase/decrease in performance in comparison with the best model in Table 1 are indicated with up/down arrows, respectively. “I” prefix refers to our substudy.