WDYT @yfarjoun @ldgauthier ?

I've never used the Picard one. I'm under the impression that the Picard version actually removes variants from the output, which I am against.

OK. I looked at the code. They do the same thing. not removing variants,
but rather apply HARD filters to variants.

The Picard version was written since GATK was too slow....

On Wed, Dec 2, 2015 at 1:59 PM, ldgauthier notifications@github.com wrote:

I've never used the Picard one. I'm under the impression that the Picard
version actually removes variants from the output, which I am against.

—
Reply to this email directly or view it on GitHub
#1215 (comment)
.

👍 for fast.

thanks for clarifying. I'll compare perf of GATK4 and report it here.

It's true - FilterVCF is faster when filtering genotypes. Here I took 1 small vcf file with 2535 samples and 12632 lines (675M)

I'm closing this then. Unclear if we want to invest in speeding up VariantFiltration given that we're moving away from vcf files at some point.

No filters: 
VariantFiltration Elapsed time: 0.20 minutes
FilterVCF Elapsed time: 1.11 minutes

MIN_DP - per-genotype filter
VariantFiltration: Elapsed time: 2.05 minutes.
FilterVCF: Elapsed time: 1.16 minutes.

MIN_GQ - per-genotype filter
VariantFiltration  Elapsed time: 2.06 minutes
FilterVCF: Elapsed time: 1.16 minutes.

One clarification - FilterVCF is not a subset of VariantFiltration, it offers one new option - filter on allele balance computed on the fly. I added #1726