Table 2.
Accuracy of SNP calling, comparing assembly-based calling with two mapping-based approaches on the same libraries’ linked read data, one each from NA12878 (L3) and NA24385 (L5).
| SNPs | True positives | False negatives | False positives | Genotype mismatch | Precision | Recall | F1 | |
|---|---|---|---|---|---|---|---|---|
| L3 (NA 12878) | Aquila (assembly only) | 3,004,501 | 38,282 | 124,074 | 4730 | 0.960 | 0.987 | 0.974 |
| FreeBayes | 3,037,504 | 5279 | 54,088 | 3501 | 0.983 | 0.998 | 0.990 | |
| Longranger | 3,040,701 | 2082 | 105,854 | 1621 | 0.966 | 0.999 | 0.983 | |
| L5 (NA 24385) | Aquila (assembly only) | 2,989,567 | 39,785 | 93,195 | 4157 | 0.970 | 0.987 | 0.978 |
| FreeBayes | 3,021,814 | 7544 | 48,477 | 3899 | 0.984 | 0.998 | 0.991 | |
| Longranger | 3,026,384 | 2974 | 104,879 | 1799 | 0.967 | 0.999 | 0.983 | |
| L5 + L6 (NA 24385) | Aquila | 2,971,237 | 58,120 | 81,926 | 18,856 | 0.973 | 0.981 | 0.977 |
The benchmark is GiaB v3.3.2. Variant counts and performance scores were generated by RTGtools/hap.py. Longranger calls were executed with “-vcmode = gatk”. For final SNP calling, Aquila combines mapping-based calls from FreeBayes with its assembly-based calls. L5 + L6 can be achieved by Aquila through a multiple-library assembly mode, which is not applicable for other tools.