Hi,
I am new to DRAGEN variant calling, previously I always use GATK best practice to call my variants. I did some research and found out that DRAGEN actually has better performance. However I have some confusion about their default hard filter (QUAL > 3, using ML calibration). In GATK, the quality threshold usually set up at Q20-Q30, I'm wondering if I need to increase my DRAGEN QUAL threshold if I want good specificity (I only have one sample). They mentioned that the GATK QUAL might be inflated, but I'm still confused if QUAL >3 is sufficient enough, since it means that only 50% probability that the variant is true.
Another question, should I pay attention to the GQ if I need high quality genotype? And what is usually the good threshold? As I know, GATK use GQ>20 for their threshold, I am not sure if I can use the same threshold for my DRAGEN data.
If there's anyone who has experience in using both tools, please help me.
Thank you!
DRAGEN uses an optimized version of GATK. See --> https://gatk.broadinstitute.org/hc/en-us/articles/4410456501915-Functional-equivalence-in-DRAGEN-GATK
You can download DRAGEN-GATK implementation here --> https://broadinstitute.github.io/warp/docs/Pipelines/Whole_Genome_Germline_Single_Sample_Pipeline/README#running-the-dragen-gatk-implementation-of-the-wgs-pipeline
Hi, thank you for the information, I already obtained my VCF file with the default filter, but I'm questioning the reliability of this default threshold, and wondering if I should perform another filtering.
Equivalence of commands in the two modes is discussed in the first link above. Did you check that?
Which DRAGEN version are you using? The filtering heavily depends on which DRAGEN version and commands you have used.
Hi, I'm using DRAGEN 4.2.7, which perform DRAGEN-ML, that's why the default filter is QUAL > 3. But I'm still wondering if this threshold is good enough for my single sample, since specificity is important here.