What is the most similar summary to VF for breakpoints that is available for breakends? I want to make a rule that determines whether a breakend is detected or not detected for each sample. I am considering using BANSR + BANRP > 0, but am not confident. Or, is QUAL more suitable? May some common end-user scenarios be added to the package vignette?
BANSR are all GRIDSS-specific fields. Unfortunately, there are no VCF-defined equivalants of the
DP fields suitable for structural variants so each caller uses it's own custom fields (or doesn't report at all). From a specifications perspective, QUAL is the only suitable field for that purpose but I don't know of any SV callers that report QUAL scores consistent with what the VCF specifications require.
What is the most similar summary to VF for breakpoints that is available for breakends?
BVF is the single breakend equivalent of
VF. As GRIDSS is (currently) the only caller to report single breakends, you should be fine using that.
Or, is QUAL more suitable?
GRIDSS uses a likelihood model that is more suitable to thresholding. It is reliant on meaningful mapq scores and in practice gives similar results to a mapq-weighted read count with a strong bias towards assembled variants (back of envelope approximation: 40 qual ~= 1 read, qual doubled if variant has assembly support).
May some common end-user scenarios be added to the package vignette?
We're intending to expand both the functionality (e.g. gene fusion prediction), and documentation of the package. This is somewhat tricky as I don't want to have too many GRIDSS-specific example.