How Do I Use Variantfiltration To Filter By Vqslod And/Or The Tranches?

Question

[Jeremy Leipzig]

After running VariantRecalibrator and ApplyRecalibration in GATK's VQSR module, I have a VCF with VQSLOD tags, as well as some labels such as "TruthSensitivityTranche99.00to99.90"

How do I use the VariantFiltration to obtain an elite VCF file? What is a reasonable minimum VQSLOD (or should I stick with tranches?)

Answer 1

[Frédéric Bigey]

Hi Jeremy,

In our study we used SelectVariant to extract variant position with VQSLOD > 4.0.

$pathToGATK \
  -T SelectVariants \
  -R reference \
  --variant inVCF \
  --select_expressions "VQSLOD >= $lodCutoff" \
  --out outName.vcf

We don't use filter like "TruthSensitivityTranche99.00to99.90".

Fred

Answer 2

[geneart$$]

I understand it has been a long time since this discussion here, but this has raised a similar question in me.

I have the tranches file generated from VQSR, however based on the tranch plot itself, I would pick 99.9 tranch to go to applyVQSR, as the TP is better in the 99.9 tranch based on the plot ,across cumulative and Tranch specific, TP. However the VQSLOD for that specific tranch is negative. So ,

1. do I need to address this issue of picking a tranch cutoff based on tranch plot or VQSLOD score before applyVQSR? or do i just go ahead with apply VQSR and then filter based on VQSLOD? Lost here, Any suggestions appreciated!
2. Also the TiTv is lower than what is suggested by GATK for a WES which is indicative of high FP.

targetTruthSensitivity,numKnown,numNovel,knownTiTv,novelTiTv,minVQSLod,filterName,model,accessibleTruthSites,callsAtTruthSites,truthSensitivity 90.00,1763732,507621,2.1238,1.0777,19.6606,VQSRTrancheSNP0.00to90.00,SNP,957871,862083,0.9000 99.00,2245012,1280942,2.1201,1.0087,4.9358,VQSRTrancheSNP90.00to99.00,SNP,957871,948292,0.9900 99.90,2423827,1416814,2.1208,1.0424,-1.6519,VQSRTrancheSNP99.00to99.90,SNP,957871,956913,0.9990 100.00,2492280,1489960,2.1202,1.0545,-22554.8245,VQSRTrancheSNP99.90to100.00,SNP,957871,957871,1.0000