We've moved!
This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!

Unified Genotyper: Maximum sensitivity

I'm seeing Mendelian errors in small but significant proportion of my unified genotyper calls. For many instances the SNP is called incorrectly as a homozygote even though there are some reads of the second allele. How can I set unified genotyper to be more sensitive ?

Best Answer


  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    The best thing you can do is switch to the HaplotypeCaller, unless your organism is not diploid.

  • BlueBlue Member

    I will do, as soon as I've uninstalled java 1.8, or there's space on the cluster. I tried HaplotypeCaller a while ago: it seems to take two weeks for ten samples, but I guess its sensitivity is better than UnifiedGenotyper. I'm still wondering if sensitivity of either caller can be altered, as a command-line argument so that even unlikely SNPs are called? I thought it might be to do with this HC argument --bandPassSigma (The sigma of the band pass filter Gaussian kernel; if not provided defaults to Walker annotated default).

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    The HC was very slow in previous versions, and this was definitely a problem for multisample calling. In the latest version, however, we have a new workflow that consists of single-sample calling with HC in GVCF mode followed by multisample joint genotyping. This gives you the same or better sensitivity as multisample calling but is much faster and can be done incrementally for projects with many samples that become available one at a time or in batches.

    There are various arguments that can tweak sensitivity somewhat, but with the new version and new workflow you shouldn't need to. Are you looking for low-frequency variants?

  • BlueBlue Member

    Thanks for the response. I'm not specifically looking for low-frequency variants.

    I'd like to increase the sensitivity of SNP-calling because:

    1) There are many homozygous Mendelian errors that would be corrected if the presence of a second allele was acknowledged by the calling algorithm.

    2) The count of unique SNPs is low compared to SNPs that occur in two or three samples, or in similar population samples.

    What are the 'various arguments' that maximise sensitivity of HaplotypeCaller ?

Sign In or Register to comment.