I'm trying to understand current workflows for SV validation and would appreciate hearing about your experiences.My understanding of the typical workflow:
Run SV caller(s) like Manta/DELLY/Lumpy on WGS data
Get VCF with hundreds to thousands of predicted SVs
Need to figure out which are real vs. artifacts
Manually review candidates in IGV by looking at read support
Questions:
Is this actually how you do it, or am I missing something?
For those doing manual review: roughly how long does it take to validate one sample's worth of SV calls?
Do you review all calls or filter/prioritize first? If prioritizing, what criteria do you use?
Are there good tools for bulk SV validation that I'm not aware of? (I've seen svviz and samplot but they seem geared toward individual SVs) For clinical/production workflows: is manual review still the standard, or has this been automated?
I'm exploring whether better tooling could help with the filtering/prioritization step (auto-scoring by evidence strength, interactive review interface). Would be useful to know if this is actually a pain point or if existing solutions work well enough.Thanks for any insights.
•
written
by
mustafa.kpn •
0 answers
No answers yet.
Log in to answer this question.
Maybe this is just my experience, but I only manually inspected IGV when absolutely necessary. Most callers give you reasonable information about the things you'd look for in an IGV track. But yes, a lot of filtering to try and identify true positives.
Something that was starting to occur when I last did an SV analysis was using two callers with different underlying identification mechanisms and only accepting SVs found in both. This is quite stringent and you can lose quite a lot of data, but it generally meant a higher quality dataset.
That said, I worked in fish genomics at the time, and it's been a few years since I did an SV analysis so the field might be in a different place now.