HaplotypeCaller: GVCF-Mode or VCF for variant calling of single sample?

Question

[DavidStreid]

Hi,

GATK recommends GVCF-mode in its Best Practices for Germline SNP/inDel discovery. However, for a single sample and not cohort analysis, does GVCF mode offer any other benefit for analysis (e.g. improved precision, sensitivity) other than speed/efficiency?

i.e.

For a single sample, why use GVCF-mode,

HaplotypeCaller ... --emit-ref-confidence GVCF -O sample.g.vcf
GenotypeGVCFs ... --variant sample.g.vcf -O sample.vcf

instead of outputting the VCF directly

HaplotypeCaller ... -O sample.vcf

Thank you - any advice or notes from personal experience would be appreciated.

Best,
David

[LChart]

There is really no point in creating a GVCF and throwing it away; however I would argue that the GVCF itself is preferable to the standard VCF for single samples, as it allows you to answer questions like "Could my sample have <variant X>"? You can look at the GVCF to see if the site is strongly homozygous reference, or poorly covered; which you cannot do with the standard VCF; and the genotypes at non-reference locations should be identical.

[Jeremy Leipzig]

right, the GVCF will distinguish reference from no-call. in the VCF that's lost. if the work is truly a single-sample then yes, skipping a step is ok. but most everything revolves around other samples.

[DavidStreid]

Makes sense - thank you, LChart & Jeremy!