Centromeric DEL detected by Manta and visible in coverage, but missing from gnomAD-SV

Structural Variation
1

Hello,

I am analyzing structural variants in a WGS cohort and have encountered a deletion that raises some questions regarding the gnomAD-SV filtering process.

The variant is a DEL on chr11:54,756,310–54,778,113 (GRCh38), which is present in approximately 50% of individuals in my cohort.

What puzzles me is the following:

  • The SV was called with Manta, which is also one of the callers used in the gnomAD-SV pipeline.

  • In the gnomAD Genome Browser, the coverage track clearly shows evidence consistent with this deletion in the same region:
    gnomAD

  • However, I cannot find a corresponding DEL in the gnomAD-SV database.

To investigate this, I reviewed the gnomAD-SV publication and supplementary materials and checked several possible explanations:

  • The region does not appear in the LUMPY exclusion blacklist (ceph18.b37.lumpy.exclude.2014-01-15.bed).

  • The deletion lies within the centromeric region of chromosome 11, which may be relevant.

  • Interestingly, I can still see INV and CPX calls reported in the same region in the browser, suggesting that SVs in this region are not systematically excluded.

My current hypothesis is that the deletion was initially detected during the gnomAD-SV analysis but was subsequently removed by one of the filtering or quality-control steps. However, I have not been able to identify which filter would explain this specific case.

Could someone clarify whether:

  1. SVs overlapping centromeric regions are subject to additional filtering criteria?

  2. A common deletion such as this could be excluded despite clear coverage evidence?

  3. INV/CPX events and DEL events are treated differently in these regions?

  4. There are known reasons why a deletion visible in the coverage data would not appear in the released gnomAD-SV callset?

Any insight into the relevant filtering steps or annotations would be greatly appreciated.

Thank you.

2

Hi @Veronique_Geoffroy, thanks for your interest in the SV data.

You’re on the right track - this variant falls in an interval that was excluded from the SV dataset because it is in a centromeric region. We have made changes to our methods since the Collins et al. 2020 paper, so while our v4 manuscript is in preparation, the best resource for up-to-date information on SV discovery methods is our GATK-SV technical documentation website. In particular, there is a page describing our reference resource files, including various exclusion lists that were applied to hard-filter variants in difficult-to-call regions (such as highly repetitive regions or regions with unreliable coverage - generally that does include centromeres and telomeres). Looking at these updated exclusion lists, you will see that the region you referred to is included in both the depth_exclude_list and the pesr_exclude_list, so depth-based CNVs overlapping this region were removed, and PE/SR-based SVs (including those from Manta) with any breakpoint inside this region were removed. As you noted, there are still a handful of SVs overlapping this region in the browser, but that is because they are from PE/SR-based callers and merely span this region, with breakpoints outside of the excluded intervals.

3

Thank you very much for your clear and detailed reply.

I really appreciate it.