Table 1. Performance of multiple variant callers based on the ICR142 dataset.
Performance metrics were calculated as: Sensitivity = TP/(TP+FN), Specificity = TN/(TN+FP) or False detection rate = FP/(FP+TP), where TP = true positive sites; TN = true negative sites; FP = false positive sites; FN = false negative sites as described in Methods. The ICR142 dataset was generated using the Illumina TruSeq exome.
| Variant type | BWA + GATK | OpEx (Stampy
+ Platypus) |
Stampy +
DeepVariant |
|
|---|---|---|---|---|
| Sensitivity | Overall | 404/416 (97%) | 391/416 (94%) | 405/416 (97%) |
| Base substitutions | 123/123 (100%) | 118/123 (96%) | 123/123 (100%) | |
| Indels | 281/293 (96%) | 273/293 (93%) | 282/293 (96%) | |
| Specificity | Overall | 266/288 (92%) | 279/288 (97%) | 270/288 (94%) |
| Base substitutions | 39/41 (95%) | 39/41 (95%) | 35/41 (85%) | |
| Indels | 227/247 (92%) | 240/247 (97%) | 235/247 (95%) | |
| False detection rate | Overall | 22/426 (5%) | 9/400 (2%) | 18/423 (4%) |
| Base substitutions | 2/125 (2%) | 2/120 (2%) | 6/129 (5%) | |
| Indels | 20/301 (7%) | 7/280 (2%) | 12/294 (4%) |