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!

Multi-tumor variant calling with Mutect2 (GATK v4.1.3.0) weakens sensitivity

amjaddamjadd FinlandMember ✭✭

This observation is very clear when one or more of the tumor samples have very low tumor content or no tumor content.

So for example if we have tumor1 and tumor2 with high tumor content and tumor3 with low tumor content, using Mutect2 on tumor1 and tumor2 will produce a lot more variants than when doing the calling on tumor1, tumor2 and tumor3 together.


  • akovalskakovalsk Member, Broadie, Moderator admin


    Thank you for your question! If tumor1 and tumor2 have much less depth than tumor3 you could lose some sensitivity, although would expect only on calls that were questionable to begin with.

    Are you applying any filtering?

  • amjaddamjadd FinlandMember ✭✭

    Hi @akovalsk
    Thank you for your answer. My observation is of course based on the final calls after FilterMutectCalls.

    But I am seeing this mainly in cases where one or more of the tumor samples are impure (almost normal like), as if they were diluting the signal. This of course could mean that Mutect2 underestimates heterogeneity in multi-sample calling, but that I cannot evaluate in my samples.

    I think that a multi-tumor somatic variant caller should emit a variant if it passes the detection limit in any of the tumor samples, not only when it passes the joint detection limit.

  • akovalskakovalsk Member, Broadie, Moderator admin

    Hi @amjadd thank you for elaborating. Could you share with us some of the sites in question? If you wouldn't mind sharing for context a few sites that were dropped, and a few that showed up in both vcfs, it may help us answer.

  • amjaddamjadd FinlandMember ✭✭

    Hi @akovalsk
    Sorry for the late reply. Here is an ultimate example of an oncogenic mutation with thousands of cosmic occurrences. The mutation was detected by single-sample calling (top panel), but not when the three samples were analyzed jointly.
    The read counts are from top to bottom: 2385,15; 1611,2; 3831,1.
    Note that the site was not emitted by Mutect2 when joint calling.
    Mutect2 command:

    gatk4 Mutect2 -R $reference -I $normal -I $tumor1 -normal $normalName  -O $vcf --max-reads-per-alignment-start 0 --pcr-indel-model HOSTILE --bam-output $outbam --germline-resource $gnomad --panel-of-normals $pon -L $targets -ip 300 --f1r2-tar-gz $f1r2

  • akovalskakovalsk Member, Broadie, Moderator admin

    Hi @amjadd thanks very much!

    Currently, Mutect2 does effectively dilute the signal as you describe - the impurity of your tumor2 and tumor3 samples obscures the variant seen in tumor1. Because all three tumors are grouped together, the detection limit is not applied per sample.

Sign In or Register to comment.