Should I Separate The Snp And Indel When Performing The Gatk Vqsr?

Question

[ugoodlfy]

Hi All, I have used GATK UnifiedGenotyper to generate a raw.vcf file. Now I want to use GATK VQSR to get a more accurate result ,and I follow this protocol:

• snp.model <- BuildErrorModelWithVQSR(raw.vcf, SNP);
• indel.model <- BuildErrorModelWithVQSR(raw.vcf, INDEL);
• recalibratedSNPs.rawIndels.vcf <- ApplyRecalibration(raw.vcf, snp.model, SNP);
• analysisReady.vcf <- ApplyRecalibration(recalibratedSNPs.rawIndels.vcf, indel.model, INDEL);

I wanna know will it be better if I seperate the SNP and INDEL when I perform VQSR, like this:

• SNP.raw.vcf , INDEL.raw.vcf <- Seperate_SNP_INDEL(raw.vcf);
• snp.model <- BuildErrorModelWithVQSR(SNP.raw.vcf, SNP);
• indel.model <- BuildErrorModelWithVQSR( INDEL.raw.vcf, INDEL);
• SNP_analysisReady.vcf <- ApplyRecalibration(SNP.raw.vcf, snp.model, SNP);
• INDEL_analysisReady.vcf <- ApplyRecalibration(INDEL.raw.vcf, INDEL.model, SNP);

Thanks a lot !

Answer 1

[Brad Chapman]

GATK best practices recommend running VQSR separately on SNPs and indels:

The UnifiedGenotyper uses a fundamentally different likelihood model when calling different classes of variation and so therefore the VQSR must be run separately for SNPs and INDELs to build separate adaptive error models

From GATK v4 best practices (although this was also true for earlier versions). See phase III, section 2:

http://www.broadinstitute.org/gatk/guide/article?id=1186

[Alex Paciorkowski]

Removed my answer, since Brad's is much more obviously correct. It's interesting how separating SNPs and INDELs are not even mentioned in the documentation I've been following: http://gatkforums.broadinstitute.org/discussion/39/variant-quality-score-recalibration

Looks like they ought to be separated at this stage after all.

[Brad Chapman]

Alex, definitely agreed. The VQSR documentation page is SNP specific, but that's not explicitly stated.