search

PapenfussLab / StructuralVariantAnnotation

R package designed to simplify structural variant analysis
GNU General Public License v3.0
69 stars 14 forks source link

How to use single breakend to increase detection sensitivity for insertions #33

Open tgong1 opened 3 years ago

tgong1 commented 3 years ago

Hi Daniel,

I have some questions regarding how to use single breakend calls to increase calling sensitivity, especially for insertions. The version of GRIDSS I have used was 2.8.3.

Q1. This is what I'm doing now to obtain SV types. I'd like to confirm that "breakpointRanges" will remove all single breakend calls? vcf = readVcf("gridss.vcf") ###the output vcf from gridss_somatic_filter.R gr = breakpointRanges(vcf) svtype <- simpleEventType(gr) ### following simple-event-annotation.R

Q2. Is this the right way to obtain single breakend calls? How I classify them into SV types? Can I see all of them as insertions detected? begr = breakendRanges(vcf)

Thank you for your time and help. Tingting

DarioS commented 3 years ago

This could be moved to StructuralVariantAnnotation Issues by the moderator.

d-cameron commented 3 years ago

breakpointRanges will remove all single breakend calls?

By default, it only reports breakpoint calls.

Q2. Is this the right way to obtain single breakend calls?

Use breakendRanges() for single breakend calls (or unpartneredBreakends=TRUE - they do the same thing).

How I classify them into SV types?

Fundamentally, GRIDSS is a breakpoint (/single breakend) detection tool. It does not classify SVs.

Can I see all of them as insertions detected?

Not necessarily. An insertion will have two single breakends at the same/nearby position in opposite orientation. LINE integrations are frequently reported as a breakpoint to the donor site and a single breakend to poly-A sequence in the opposite orientation. MSI expansions also get frequently reported as single breakends. Viral integrations also show up as single breakends (but are more reliably detected with VIRUSBreakend)

If you want to do SV classification, then I strongly suggest using LINX classifications (it now supports tumour-only runs). LINX actually does perform SV classification and has pretty good event resolution capabilities.