DRAGEN-GATK and joint genotyping

Follow

Le Qi

• September 26, 2023 02:09
• Edited

I have a general question about the new DRAGEN-GATK pipeline and how it interacts with joint genotyping step. I understand that the DRAGEN-GATK is currently supporting only single sample mode, and only hard filtering is required on QUAL. I'm wondering how different are the gVCF generated in this pipeline as compared to the traditional GATK pipeline, particularly regarding the info used by joint genotyping step. Is there obvious incompatibility between DRAGEN-GATK and joint genotyping? 

Thanks!

0

• 

  Ricky Magner

  • September 27, 2023 16:21

  Hi Le,

  On a technical level, the gVCFs produced by DRAGEN-GATK are still valid gVCFs produced by GATK, so they should pass through tools expecting that sort of input. I am not aware of any big studies comparing performance of joint callsets aside from some small internal analyses, but the single sample outputs were benchmarked to be functionally equivalent to DRAGEN (at least for some matching of versions on either side). Note that some of the annotations however may be a bit different between the two tools, so you might want to check if your pipeline for joint calling is sensitive to those. 

  Best,

  Ricky

  0

  Comment actions Permalink
• 

  Le Qi

  • September 27, 2023 16:51

  Hi Ricky Magner

  Thanks for your reply! I was specifically talking about the genotypeGVCFs tool from GATK. I could run the DRAGEN-GATK output gVCF through genotypeGVCFs without problems. I'm curious if the difference between VQSR used by regular GATK and hard-filtering recommended by DRAGEN makes any differences in the GATK joint genotyping pipeline results. If I understand correctly, the current GATK joint genotyping pipeline still uses VQSR. Are the improvements from incorporating DRAGEN in the single sample mode still maintained after joint genotyping?

  Sorry if my question doesn't make sense. I can't find details about how the GATK joint genotyping pipeline works, or what information it utilizes from gVCFs to make joint calls.

  0

  Comment actions Permalink
• 

  Ricky Magner

  • September 29, 2023 17:05

  Hi Le,

  I spoke with someone who did an informal analysis internally who said they said modest improvements by following the DRAGEN filtering strategy, but VETS (the VQSR replacement) is also pretty good. I think it would be your decision to see if there might be any other factors for your analysis to prefer one way or another.

  Best,

  Ricky

  0

  Comment actions Permalink

Please sign in to leave a comment.