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!

LOH detection using GATK4's somatic CNV workflow

jasonbwarnerjasonbwarner Greater Boston AreaMember
Can GATK4's somatic CNV workflow detect copy-neutral LOH events for 1) tumor-only samples with a panel of normal (PON) or 2) tumor-normal pairs with a PON? I see that ModelSegments can give a "0" call, but it seems like that is for all high-quality neutral calls.


  • sleeslee Member, Broadie, Dev ✭✭✭
    edited August 2019

    Thanks for the question, @jasonbwarner! Although the current GATK somatic CNV pipeline uses relatively sophisticated methods for segmentation/modeling of CR/AF, its calling capability is somewhat limited (and is more on par with standard methods for arrays, etc.). This is partially due to the development history of this pipeline (which was originally based on older methods developed by Broad CGA), as alluded to by the documentation for the CallCopyRatioSegments caller:

    Calls copy-ratio segments as amplified, deleted, or copy-number neutral.
    This is a relatively naive caller that takes the modeled-segments output of ModelSegments and performs a simple statistical test on the segmented log2 copy ratios to call amplifications and deletions, given a specified range for determining copy-number neutral segments. This caller is based on the calling functionality of ReCapSeg.

    If provided ModelSegments results that incorporate allele-fraction data, i.e. data with presumably improved segmentation, the statistical test performed by CallCopyRatioSegments ignores the modeled minor-allele fractions when making calls.

    I am currently working on a new tool for determining purity/ploidy, subclonal structure, and calling absolute allelic copy number, which will take as input the results from ModelSegments and Mutect2. This tool will eventually supersede CallCopyRatioSegments.

    However, until then, you might try rolling your own methods that combine the log2CR/MAF estimates output by ModelSegments with the calls from CallCopyRatioSegments to identify potential CNLOH. When the matched normal is available and the CNLOH events are large, I'd expect relatively simple methods (e.g., hard filtering) might suffice. Tumor-only calling, calling small CNLOH events, and distinguishing ROH may be more challenging.

Sign In or Register to comment.