If you happen to see a question you know the answer to, please do chime in and help your fellow community members. We encourage our fourm members to be more involved, jump in and help out your fellow researchers with their questions. GATK forum is a community forum and helping each other with using GATK tools and research is the cornerstone of our success as a genomics research community.We appreciate your help!

Test-drive the GATK tools and Best Practices pipelines on Terra

Check out this blog post to learn how you can get started with GATK and try out the pipelines in preconfigured workspaces (with a user-friendly interface!) without having to install anything.

different QUAL for invariant and variant sites


Is there any way to make an equivalent filter on variant and invariant sites based on QUAL?. I want to apply a higher QUAL filter for variant sites since they have different distributions. I was thinking of a QUAL of 30-40 for invariant sites, but I needed to decide on a similar QUAL for variant sites. I was thinking one possibility was to get the correlation of QUAL values for close variant and invariant sites (within 2bp). However, although significant, this correlation is not very high, around 0.5.

If I use a very low QUAL for variant sites I will get too many false positive alt variants, especially singletons.



  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    Hi Jose,

    FYI the current version of GATK does not emit quals for hom-ref sites, so you cannot filter them based on qual. Former versions did emit quals for hom-ref but those values were wrong; that's why that behavior was inactivated. If you're using an older version that emits quals for invariant sites, we advise you not to use them. It would be better to upgrade to the latest version and follow our Best Practices for filtering variants.

  • How recent is that version of GATK?, I have downloaded it 3 weeks ago. I am using GenomeAnalysisTK-2.2-9-g54ae978. These are the results of the vcf file that out of UnifiedGenotyper.

    Thank you.

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    Oh, I just realized you're talking about actual QUAL, I was thinking of GQ. Sorry about that.

    You should be able to do this conditional filtering by using a JEXL expression to evaluate the genotype call then applies the corresponding thresholds. Might be a bit tricky so you'll have to experiment to get it right. If that proves too tricky you can always subset your variants from your invariants and filter them separately.

  • ciprianillociprianillo Member
    edited December 2012

    Sorry, I should have used QUAL rather than qual. I'll start just sub-setting variants and invariants separately.

    Any recommendations to select a QUAL threshold for invariant sites (that are in the 10's) and one for variant sites (that tend to have much higher scores, in the 1000's)?

    Post edited by ciprianillo on
  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    No worries, you wrote it correctly; but we recently had several questions about GQ so I jumped to the conclusion that this was more of the same. My bad.

    We only really use QUAL at the level of the variant caller, to decide which sites to emit, and we tend to be very lenient. Then afterwards we employ strict filtering methods, but we use other metrics. See the Best Practices and VQSR documents for more details on that.

  • sorrywmsorrywm Member

    Just to clarify, does this mean that there is no way in GATK to get phred-scaled quality scores for invariant sites that are comparable to those for variant sites (in that they represent the probability of the call being wrong)? I am looking for something similar to the consensus quality (FQ) output by samtools/mpileup. Or perhaps abs(FQ) is exactly the same as QUAL but similarly poorly calibrated. I apologize if I'm posting this in the wrong place, but it seemed related to the previous questions.

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    I think what you're looking for is the QUAL, which works fine for SNPs.

  • scitechscitech new delhiMember

    Hello Ciprianeillo and Geraldine,
    I am also trying to get invariant sites using GATK-2.7-2 Unified Genotyper, so want to ask which parameter i should set for getting the same.
    can it be done by getting phred quality scores for invariant sites? and also how can i subset variant and invariants seperately ?

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    @scitech, have a look at the options for the UG's output mode. Note however that UnifiedGenotyper is not able to produce qual scores for invariant sites. For that, you need to use the HaplotypeCaller's reference model feature. For subsetting, have a look at the technical documentation for SelectVariants.

  • scitechscitech new delhiMember

    Thank you Geraldine

  • scitechscitech new delhiMember

    Hello Geraldine, I have got the vcf file for invariants, and i did it with taking only one exome file and a reference file, but the file at some positions giving GT ./. that means GT call cannot be made at those positions. So my question is does GT for one sample can also have missing allele information or I am getting some malformed file???

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    It is normal to have some positions with ./. in the GT field. This happens when there is no useable coverage at a given site. You may want to use DiagnoseTargets to find out how many targets in your exome are insufficiently covered.

  • newbie16newbie16 Member

    Hello Geraldine

    In an earlier post you mentioned that "UnifiedGenotyper is not able to produce qual scores for invariant sites. For that, you need to use the HaplotypeCaller's reference model feature."
    However, I am seeing the opposite. I can see QUAL for invariant sites by Unified Genotyper but not by Haplotype Caller. Could you please clarify?
    On the other hand I see that UG does not have GQ for invariant sites but HC outputs GQ for invariant sites.

  • SheilaSheila Broad InstituteMember, Broadie, Moderator admin



    Thanks for catching this. Geraldine meant GQ, not QUAL.


Sign In or Register to comment.