Table 5:
Accuracy of algorithms by broad lineage.
| Algorithm | Overall | EAI | East-Asian | Euro-American | Indo-Oceanic | M. africanum | M. bovis |
|---|---|---|---|---|---|---|---|
| MIRU-VNTRplus (*) | 44.5 [97.8] | 44.1 | 69.9 | 31.9 | 53.7 | 70.0 | 100 |
| Same, no threshold (*) | 96.9 | 88.2 | 99.8 | 98.1 | 94.4 | 90.0 | 100 |
| MIRU-VNTRplus (†) | 91.8 [97.1] | 88.0 | 97.3 | 93.8 | 77.2 | 93.3 | 100 |
| Same, no threshold (†) | 96.3 | 88.0 | 98.6 | 98.2 | 90.1 | 93.3 | 100 |
| TB-Insight | 96.3 | 82.4 | 97.3 | 98.2 | 96.3 | 90.0 | 100 |
| TBminer - MIRU-VNTRplus | 96.3 | 79.4 | 98.6 | 98.2 | 96.3 | 90.0 | 100 |
| TBminer - Expert | 95.8 | 82.3 | 98.6 | 98.2 | 92.6 | 80.0 | 100 |
| TBminer - MIRU-VNTRplus (‡) | 95.8 | 82.4 | 94.5 | 98.2 | 96.3 | 90.0 | 100 |
| TBminer - Expert (‡) | 95.0 | 79.4 | 94.5 | 97.8 | 94.4 | 90.0 | 100 |
| RuleTB | 95.0 | 88.2 | 95.9 | 96.0 | 94.4 | 90.0 | 100 |
| StackTB | 97.0 | 88.2 | 98.6 | 98.2 | 96.3 | 90.0 | 100 |
(*)
= the MIRU-VTNRplus database is used for training;
(†)
= the training subset of the entire data is used for training;
‡
= only the 15 most discriminative VNTR loci are used for the prediction. The numbers in square brackets indicate the accuracy of only the assignments actually made (excluding NAs).