Small variant calling and structural variant calling with Dragen is possible now.
| WES (NA12878_58) | freebayes | dragen v3.8.4 | dragen v4.0.3 |
|---|---|---|---|
| all | 128106 | 140852 | 140724 |
| AF<1% | 5573 | 6277 | 6179 |
| AF<1%, NGSD<20 | 1400 | 2919 | 2857 |
| AF<1%, NGSD<20, high/moderate/low impact | 356 | 410 | 392 |
| recessive stringent | 38 | 62 | 55 |
| recessive relaxed | 69 | 111 | 99 |
| dominant stringent | 10 | 9 | 9 |
| dominant relaxed | 162 | 185 | 174 |
| dominant relaxed + pheno | 72 | 198 | 197 |
| rank1: | chr11:6617154 C>T (8.00) | chr11:6617154 C>T (8.00) | chr11:6617154 C>T (8.00) |
| overlap with freebayes calls (Agilent v7 ROI) | 97.05% + 1.27% GT diff | 96.94% + 1.26% GT diff | |
| overlap with freebayes calls (Agilent v7 ROI, only high conf region) | 99.31% + 0.18% GT diff | 99.27% + 0.17% GT diff |
| WGS (NA12878_45) | freebayes | dragen v4.0.3 |
|---|---|---|
| all | 365620 | 373779 |
| AF<1% | 278628 | 284172 |
| AF<1%, NGSD<20 | 43214 | 92845 |
| AF<1%, NGSD<20, high/moderate/low impact | 406 | 478 |
| recessive stringent | 46 | 100 |
| recessive relaxed | 89 | 167 |
| dominant stringent | 26 | 100 |
| dominant relaxed | 189 | 229 |
| dominant relaxed + pheno | 1140 | 1896 |
| rank1: | chr11:6617154 C>T (8.50) | chr11:6617154 C>T (8.50) |
| overlap with freebayes calls (Agilent v7 ROI) | 95.21% + 1.81% GT diff | |
| overlap with freebayes calls (Agilent v7 ROI, only high conf region) | 98.05% + 0.91% GT diff |
Conclusion: Small variant calls are pretty similar, except for artefacts in low complexity regions.
Bechmarks on GIAB sample NA12878 based on the Agilent V7 exome target region:
| SNVs | sensitivity | ppv | genotyping |
|---|---|---|---|
| WES default | 99,87 | 98,92 | 99,96 |
| WES dragen v4.0.3 | 99,89 | 99,52 | 99,98 |
| WGS default | 99,95 | 98,22 | 99,60 |
| WGS dragen v4.0.3 | 99,91 | 99,19 | 99,95 |
| InDels | sensitivity | ppv | genotyping |
|---|---|---|---|
| WES default | 95,29 | 93,72 | 96,21 |
| WES dragen v4.0.3 | 97,64 | 95,93 | 98,49 |
| WGS default | 98,24 | 97,85 | 98,67 |
| WGS dragen v4.0.3 | 99,74 | 99,15 | 99,79 |
Structural variant calling with Dragen is performed only if dragen_sv_calling is true in the settings.
| WES (NA12878_58) | freebayes | dragen v3.8.4 | dragen v4.0.3 |
|---|---|---|---|
| all | 201 | 198 | 238 |
| nospecial, filter PASS | 82 | 75 | 86 |
| nospecial, filter PASS, qual>=100 | 72 | 72 | 86 |
| nospecial, filter PASS, qual>=100, PE read depth>=5 | 50 | 46 | 44 |
| nospecial, filter PASS, qual>=100, PE read depth>=5, NGSD<=30 | 5 | 27 | 27 |
| nospecial, filter PASS, qual>=100, PE read depth>=5, NGSD<=30, bpd<=20 | 0 | 7 | 4 |
| exome/wgs stringent | 5 | 27 | 27 |
| exome/wgs relaxed | 15 | 97 | 129 |
| SVs MANTA (qual>100): | 146 | 146 | |
| SVs DRAGEN (qual>100): | 172 | 201 | |
| overlap (pos): | 93 | 92 | |
| overlap (pos+type): | (type diff) 63 | (type diff) 62 | |
| overlap (pos+type+GT): | 63 | 60 |
| WGS (NA12878_45) | freebayes | dragen v4.0.3 |
|---|---|---|
| all | 10992 | 13699 |
| nospecial, filter PASS | 8236 | 12298 |
| nospecial, filter PASS, qual>=100 | 7734 | 12074 |
| nospecial, filter PASS, qual>=100, PE read depth>=5 | 3725 | 4908 |
| nospecial, filter PASS, qual>=100, PE read depth>=5, NGSD<=30 | 65 | 579 |
| nospecial, filter PASS, qual>=100, PE read depth>=5, NGSD<=30, bpd<=20 | 43 | 151 |
| exome/wgs stringent | 116 | 621 |
| exome/wgs relaxed | 765 | 3272 |
| SVs MANTA (qual>100): | 8585 | |
| SVs DRAGEN (qual>100): | 12643 | |
| overlap (pos): | 7197 | |
| overlap (pos+type): | 6941 | |
| overlap (pos+type+GT): | 6797 |
Conclusion: Structural variant calls differ quite a bit. Dragen calls more variants, but is more conservative when not sure about the type (reports BND instead of DEL/DUP from Manta). It also reports small tandem duplications (< 1000 bases) as insertions!
GutHub issue: #132
Benchmarks were performed in the folder: /mnt/storage3/users/ahsturm1/Sandbox/2022_09_07_dragen_vc/