split-vep only prints first numerical value

[sbaross]

I have a VCF of structural variants which has been annotated with overlapping gnomAD v4.1 structural variants using VEP custom annotations, and I'm using split-vep to convert this data into a tab-delimited format for further analysis.

Where my variants overlap with multiple gnomAD SVs, split-vep will print all the IDs delimited with &, but only the first allele frequency. For example:
bcftools +split-vep -f "[%gnomAD %gnomAD_AF]"
gives
gnomAD-SV_v3_DEL_chr1_90cb8e69&gnomAD-SV_v3_DEL_chr1_48a6a36f&gnomAD-SV_v3_DEL_chr1_58a8b87a&gnomAD-SV_v3_DEL_chr1_c80d6f1c 4.1e-05

However if I specify the AF field as string first I get all values as expected (which surprises me as the fields are specified as strings in the header and I've previously raised an issue here when I was struggling to override that!)
bcftools +split-vep -f "[%gnomAD %gnomAD_AF]" -c "gnomAD_AF:Str"
gives
gnomAD-SV_v3_DEL_chr1_90cb8e69&gnomAD-SV_v3_DEL_chr1_48a6a36f&gnomAD-SV_v3_DEL_chr1_58a8b87a&gnomAD-SV_v3_DEL_chr1_c80d6f1c 4.1e-05&8.7e-05&0.027837&0.003762

I'm working in a restricted research environment so can't easily share an example file, but if more information is needed to replicate the issue I can try figure something out

[pd3]

Hi, to debug this I’ll need the VCF header definition of the VEP field and the corresponding INFO consequence. You can modify the actual values to make the record biologically meaningless, just preserve the structure so I can reproduce the issue.

[sbaross]

Apologies for the delay - I've attached a sample VCF containing one variant with multiple overlapping gnomAD annotations and the header which should allow you to replicate the issue

sample.vcf.gz

[pd3]

There seems to be a mistake, the file you sent contains this

##fileformat=VCFv4.2
#CHROME POS     ID      REF     ALT     QUAL    FILTER  INFO
chr1    159678918       .       N       <DEL>   .       .       SVTYPE=DEL;SVLEN=-950;END=159679868

[sbaross]

Apologies, wrong test file! This is the correct one

test_annotated.vcf.gz

[pd3]

Thank you. I see what is happening: +split-vep does not recognize the & subdivisions, it only splits by transcript (',') and field field ('|'). Therefore, in numeric parsing mode, it only returns the numeric part of 4.1e-05&8.7e-05, ignoring whatever comes after '&'. Forcing AF into a string works around the issues.