Table 3. Comparison of algorithm performance for different CNV sizes.
Results shown for short (500–10,000 bases) / long (>10,000 bases) CNVs. True positives indicate 10% reciprocal overlap. CNV predictions not meeting overlap criteria were classified as “Other.”
| Deletion | Duplication | |||||
|---|---|---|---|---|---|---|
| Algorithm | Sensitivity | True positives | Other | Sensitivity | True Positives | Other |
| NA12878 (low coverage) | ||||||
| CNVnator | 0.11 / 0.72 | 47 / 55 | 202 / 346 | 0.03 / 0.27 | 3 / 25 | 20 / 186 |
| RDXplorer | 0.03 / 0.34 | 11 / 26 | 62 / 156 | 0.03 / 0.04 | 3 / 4 | 217 / 132 |
| GROM-RD | 0.37 / 0.84 | 153 / 64 | 740 / 123 | 0.05 / 0.27 | 6 / 25 | 86 / 227 |
| NA12878 (high coverage) | ||||||
| CNVnator | 0.78 / 0.81 | 328 / 63 | 27132 / 465 | 0.09 / 0.23 | 12 / 22 | 618 / 357 |
| RDXplorer | 0.16 / 0.62 | 69 / 48 | 1418 / 232 | 0.05 / 0.15 | 7 / 15 | 595 / 199 |
| GROM-RD | 0.68 / 0.83 | 287 / 65 | 5413 / 156 | 0.15 / 0.26 | 20 / 25 | 1216 / 252 |