We are currently performing an urgent review of a BRIP1 deletion (encompassing exon 8) and was wondering if you could help us understand the quality scores for SVs in gnomAD v4.1.0.
We believe our variant overlaps with structural variant: DEL_CHR17_69E6FF13, which has been called in 4/126092 total alleles sequenced in gnomAD SVs v4.1.0. But ¾ of these variant carriers have a genotye quality score of 0-5, please can you confirm if we can be confident with these calls or if these should be excluded from our allele frequency calculation? Note this variant has an overall filter of pass but we are concerned with these 3 calls.
This is currently making a huge difference as to whether we report our variant, so would really appreciate a response as soon as you can.
I checked this variant, and it turns out that the three samples with genotype quality (GQ) scores of 0–5 are all from PCR+ libraries, whereas the one sample with a high GQ score is from a PCR– library. I manually reviewed the depth profiles for all carriers and observed clear sequencing-depth support in the PCR– sample with high GQ. In addition, this sample also shows supporting evidence from paired-end and split-read signals, so I consider the deletion to be real for this carrier. However, the PCR+ samples with low GQ scores show questionable depth profiles. To be conservative, I would not treat their genotypes as manually validated.
I also want to provide some context: this 5.3 kb deletion is above—but close to—our 5 kb size cutoff, above which our pipeline allows variants supported solely by sequencing depth to be retained as PASS if depth is the only evidence that supports the non-reference genotype. Unfortunately, depth profiles are generally much noisier in PCR+ libraries than in PCR– libraries, which leads to higher false-positive rates in PCR+ samples, particularly for relatively small CNVs in the 5–20 kb range.
I hope that helps, but please let me know if there are additional questions that I can help with.
Thank you for all the information- this has been really useful and I have relayed this to our senior scientists to review. I did raise this with our lead bioinformatician and his thoughts were similar, that he would be skeptical that these 3 poor quality calls are real.
One other thing that I am hoping you can confirm for this variant and for the gnomAD v4.1.0 SVs in general. Does this include UK Biobank data? Just trying to ascertain if we are able to determine if this variant is absent from the UK Biobank.