clarity on what is being filtered

[TheTetractys]

Hello!
I am working on performing some validation on technical replicates (DRAGEN SV vs DRAGEN SV) and I noticed that a large portion of the variants in the vcfs are being filtered out.
Example:
2025-12-04 11:01:01,811 [INFO] 27302 chunks of 57483 variants Counter({'base': 21722, 'comp': 19263, '__filtered': 16498})

I was using passonly false, as i was interested in seeing performance with all variants. Are the '__filtered' variants that have been collapsed down (duplicative calls)? What other parameters am I using that are filtering so many variants out?

After a bit more digging it appears that partially assembled INS will be given a SVLEN of 0, so these were all filtered out from consideration in bench. This could be overcome with --sizemin 0 parameters, is this the best way to handle this?

Thank you!

base"/s_XXXX__rep1.sv_annotated_final.vcf.gz" comp"/s_XXXX__rep2.sv_annotated_final.vcf.gz" output"/truvari_s_XXXX_r1vr2_all_single" includebednull extend0 debugfalse referencenull refdist50 pctseq0.7 pctsize0.7 pctovl0.9 typeignorefalse no_rollfalse chunksize1000 bSample"s_1235260B__rep1" cSample"s_1235260B__rep2" dup_to_insfalse bnddist100 sizemin50 sizefilt30 sizemax50000 passonlyfalse no_reffalse pick"single" ignore_monreftrue check_multitrue check_monreftrue no_single_bndtrue write_resolvedfalse decomposetrue short_circuitfalse skip_gtfalse max_resolve25000

[ACEnglish]

Hello,

The filtered variant counts are SVs skipped for failing the FILTER field (but only with --passonly) as well as SVs outside the --sizemin / --sizefilt and --sizemax boundaries and a few others. See here for details on filtering. As you described, if the partial insertions have an SVLEN of 0, those were being filtered. See here for details of how SV size is determined.

Have a great day,
~/Adam English