Question: using minda to match new SV calls to existing panels

[ajw2329]

Hello and thanks for the great tool!

I'm interested in being able to compare newly generated SV calls (generated, for instance, using severus) to existing panels of SVs. My current specific use case is the annotation of long read germline SV calls using 1KGP-ONT frequencies.

This doesn't exactly fall into either the truthset or ensemble options, but I imagine minda could be useful nonetheless. I've experimented with the ensemble tool, including my germline panel as one of the VCFs as a start -- this seems to work OK but I wanted to check my thinking. Any advice/recommendations would be much appreciated!

[mikolmogorov]

Hi @ajw2329

Thanks for the feedback! I'll let @asherbryant answer wrt to Minda specifically, and it could be potentially useful! I imagine it will require some work on the code.

In the meantime, I also wanted to highlight a recent tool that supports 1KG long-read annotations: https://github.com/jmonlong/variant_annotation_wf (from this manuscript - https://www.cell.com/ajhg/fulltext/S0002-9297(25)00002-3). Maybe it could be helpful as well?

[ajw2329]

This looks great -- thanks very much for the suggestion!

[eesiribloom]

@ajw2329 did you have success annotating your long read germline SV calls with 1KGP-ONT ? If so how did you go about it?

[ajw2329]

@eesiribloom I did have success running minda ensemble on my calls + 1KGP-ONT, which I think is a reasonable approach. For my application it is also important to independently check that the predicted protein consequences are the same for both the novel call and the nominally equivalent panel event, in case the two are not already identical at the base-level.

I was not able to try out the variant_annotation_wf linked above -- IIRC at the time there were some missing annotation prep instructions -- looking at the repo now it looks like they've updated that info ( annotation prep script ) so I might give it another shot. Under the hood that is using sveval -- I'm not sure if that approach is using the same BND-decomposition approach that makes minda attractive.

Side note: no approaches I've tried (except for the simplest event type-agnostic overlap approach) are catching scenarios where domestic insertions are represented as BND quartets in one VCF and insertions in another.