How to make sniffles do a better job in splitting insertions at the same location into two het calls?

[ymcki]

I am running sniffles with HG002 data from ONT on NIST CMRG v1.00 benchmark. This benchmark has 216 SVs.

I find that most of the false negatives are due to failure of sniffles to call two different insertions at the same position as two heterozygous calls and simply call them as a homozygous, e.g. this one is called as a homo 304bp insertion
chr3 45890270 Sniffles2.INS.2C7S2 N <INS> 60 PASS IMPRECISE;SVTYPE=INS;SVLEN=304;END=45890270;SUPPORT=43;COVERAGE=43,43,43,43,42;STRAND=+-;AF=1.000;STDEV_LEN=31.472;STDEV_POS=0.000;SUPPORT_LONG=0 GT:GQ:DR:DV 1/1:60:0:43

In the above case, the correct call should be 245bp insertion on one haplotype and 309bp on the other.

I found that when the sizes of insertions differ more significantly, they can be called as two heterozygous calls, e.g.
chr9 137102263 Sniffles2.INS.8DCS8 N <INS> 60 PASS PRECISE;SVTYPE=INS;SVLEN=267;END=137102263;SUPPORT=18;COVERAGE=33,33,33,33,34;STRAND=+-;AF=0.545;STDEV_LEN=0.000;STDEV_POS=0.000;SUPPORT_LONG=0 GT:GQ:DR:DV 0/1:60:15:18
chr9 137102263 Sniffles2.INS.8DDS8 N <INS> 60 PASS PRECISE;SVTYPE=INS;SVLEN=1670;END=137102263;SUPPORT=14;COVERAGE=33,33,33,33,34;STRAND=+-;AF=0.424;STDEV_LEN=2.493;STDEV_POS=9.054;SUPPORT_LONG=0 GT:GQ:DR:DV 0/1:60:19:14

So what parameters can I change to make the latter happens more often? Thanks a lot in advance.

[fritzsedlazeck]

Dear @ymcki
we are just releasing a new version. The code version is now alreayd online for a few days.
This should improve the behavior to not over merge INS as the sequence identify is taken into consideration.
thanks
Fritz