Table 19 HELLO without indel realignment: results for HG003 chr21-22
From: HELLO: improved neural network architectures and methodologies for small variant calling
| Â | DeepVariant | HELLO | ||||
|---|---|---|---|---|---|---|
Precision | Recall | Errors | Precision | Recall | Errors | |
a. Illumina indel | ||||||
30x | 0.990523 | 0.983499 | 361 | 0.992667 | 0.981609 | 356 |
b. Illumina SNV | ||||||
30x | 0.994724 | 0.992616 | 1088 | 0.997012 | 0.992349 | 914 |
c. Hybrid indel | ||||||
20 × 15x | 0.988237 | 0.989823 | 308 | 0.991680 | 0.989096 | 268 |
d. Hybrid SNV | ||||||
20 × 15x | 0.998732 | 0.997512 | 323 | 0.998581 | 0.997442 | 342 |