RTG vcfeval: number of variants reported by vcfeval is not equal to actual number of variants.

Question

[c.clarido]

Hello community,

I am running the rtg-tool vcfeval as follow:

rtg vcfeval  
--baseline baseSnps.vcf.gz \
--calls calledSnps.vcf.gz \
--output rtg-results \
--template ref.sdf \
--sample "Macrogen-48-0511,48-0511" \
--output-mode "split"

The ref.sdf is generated from GRCh37. Both VCF files were based on this build.

I noticed however, that the number of variants reported by this tool is a lot less that the actual number of variants found in both vcf files.

baseSnps.vcf.gz contains 378285 variants calledSnps.vcf.gz contains 1125224 variants

However, looking at the summary:

Threshold  True-pos-baseline  True-pos-call  False-pos  False-neg  Precision  Sensitivity  F-measure
----------------------------------------------------------------------------------------------------
     None            3508339        3509333     525719      52825     0.8697       0.9852     0.9238

TPbase + FN = 3508339 + 52825 = 3561164
TPcall + FP = 3509333 + 525719 = 4035052

Is there something, I'm missing?

[Kevin Blighe]

I would contact Len Trigg and RTG directly. He is a user on Biostars but does not appear that often. https://www.realtimegenomics.com/company/len-trigg

Kevin

Answer 1

[Len Trigg]

By default, vcfeval excludes variants that are not PASS or . in the VCF FILTER column (you can use --all-records if you want to include these).

There are other variant classes that are out-of-scope for the matching process, such as SVs, very long indels, non-variant records (i.e. GT is 0/0), etc.

If you run vcfeval in annotation mode (--output-mode annotate), the output VCFs will contain every record present on the input, with additional annotation fields indicating status, so you could see specifically which variants are being ignored.

[Kevin Blighe]

Thanks for stopping by Len. Trust that all is well.