truvari collapse- inversions and duplications

[Madelinehazel]

Hello,

Thanks for this tool! I am using truvari to collapse SVs and I'm using --pctovl 0.5 to specify a 50% reciprocal overlap between deletions. Is there a reason that this isn't applied to inversions and duplications? From what I can tell, inversion and duplication matching is only restricted by the --refdist and --pctsize (as with insertions).

Thanks,
Madeline

[ACEnglish]

Hello,

Truvari was initially built for use with sequence-resolved breakpoint-exact DEL and INS, which meant the INS had no span over the reference. Only recently (f0a6088) in v3.1 did we add a boundary inflation for INS. But I did accidentally leave the line which restricts matching to DEL.

I'll fix this and have it slated for the next release.

[ACEnglish]

Please install from develop (see wiki for details] if you'd like to use the new fix now. I've been planning to cut a v3.2.0 release on Apr 1st, so your changes will be available to install via pip at the end of the week.

[Madelinehazel]

Hello,

Thank you for the quick response! I don't think I'll need reciprocal overlap for matching INS, as the --refdist and --pctsize should be sufficient for my purposes. Will the next release allow the reciprocal overlap to be applied to INV and DUP collapsing?

Thanks,
Madeline

[ACEnglish]

The change is that --recovl is applied to all SV pairs during Matching. So yes, it will be applied to INV and DUPs as well. However, it will also be applied to INS. With the INS boundary changes mentioned above, it seems more faithful to the parameter description and general user expectations for the parameter to be applied to all SV pairs.

I will note that in our paper we found that reciprocal overlap isn't a very robust matching criteria for sequence-resolved, breakpoint exact SVs, which is why the default remains at 0.

Interestingly, 402 SV pairs have ≤5% reciprocal-overlap but ≥70% sequence and size similarity. These pairs may indicate alignment ambiguities across repetitive regions (e.g. left-shift vs. right-shift).

[Madelinehazel]

Got it, thank you! Looking forward to the next release. Point taken re: robustness of matching criteria, though- will take this into consideration.