Variant confidence normalized by unfiltered depth of variant samples (QD)
Category Variant Annotations
Overview
Variant confidence normalized by unfiltered depth of variant samplesThis annotation puts the variant confidence QUAL score into perspective by normalizing for the amount of coverage available. Because each read contributes a little to the QUAL score, variants in regions with deep coverage can have artificially inflated QUAL scores, giving the impression that the call is supported by more evidence than it really is. To compensate for this, we normalize the variant confidence by depth, which gives us a more objective picture of how well supported the call is.
Statistical notes
The QD is the QUAL score normalized by allele depth (AD) for a variant. For a single sample, the HaplotypeCaller calculates the QD by taking QUAL/AD. For multiple samples, HaplotypeCaller and GenotypeGVCFs calculate the QD by taking QUAL/AD of samples with a non hom-ref genotype call. The reason we leave out the samples with a hom-ref call is to not penalize the QUAL for the other samples with the variant call.
Here is a single sample example:
2 37629 . C G 1063.77 . AC=2;AF=1.00;AN=2;DP=31;FS=0.000;MLEAC=2;MLEAF=1.00;MQ=58.50;QD=34.32;SOR=2.376 GT:AD:DP:GQ:PL:QSS 1/1:0,31:31:93:1092,93,0:0,960
QUAL/AD = 1063.77/31 = 34.32 = QD
Here is a multi-sample example:
10 8046 . C T 4107.13 . AC=1;AF=0.167;AN=6;BaseQRankSum=-3.717;DP=1063;FS=1.616;MLEAC=1;MLEAF=0.167;QD=11.54 GT:AD:DP:GQ:PL:QSS 0/0:369,4:373:99:0,1007,12207:10548,98 0/0:331,1:332:99:0,967,11125:9576,27 0/1:192,164:356:99:4138,0,5291:5501,4505
QUAL/AD = 4107.13/356 = 11.54 = QD
Caveat
This annotation can only be calculated for sites for which at least one sample was genotyped as carrying a variant allele.
Related annotations
- Coverage gives the filtered depth of coverage for each sample and the unfiltered depth across all samples.
- DepthPerAlleleBySample calculates depth of coverage for each allele per sample (AD).
GATK version 4.0.6.0 built at 25-39-2019 01:39:46.
0 comments
Please sign in to leave a comment.