ALLREADS flag discrepancy

[tfwillems]

I believe there is a discrepancy between the ALLREADS format field and the GB format field in the VCF allelotyper output. The length differences reported by the two fields should match but it appears that they're off by the length of one repeat unit. For instance, given the format identifier ALLREADS:AML:DP:GB:Q:GT:STITCH, here are some VCF entries with a discrepancy

1. For a dinucleotide locus:
   0|2:0.673853/0.673853:2:-2/-2:0.107933:10/10:0
   6|2:0.673006/0.673006:2:4/4:0.107592:12/12:0
2. For a tetranucleotide locus:
   -15|1:0.69445/0.69445:1:-19/-19:0.145835:1/1:0
   -7|2:0.830356/0.830356:2:-11/-11:0.262465:3/3:0

These discrepancies are ubiquitous and I believe affect every call. Is this indeed an error, or am I misunderstanding the way the field is specified?

[mgymrek]

Which version are you using? I don't see this issue with the latest version on github (3.0.3.22-c37a)

[tfwillems]

I observed this issue with the latest version released on the website (3.0.3)

[tfwillems]

Is it possible this issue is limited to STRs with a name in the last column of the reference bed? I observed this issue when I was inspecting calls for the annotated markers specifically, but perhaps it doesn't affect the other loci?

[mgymrek]

seems this only happens when using the old noise model (illumina_v2.0.3) with lobSTR3. We need to generate a new noise model for not-PCR free data with the new lobSTR3 and to update the website to say to only use newer noise models with lobSTR3.