AlleleBalance changes

bgrenierbgrenier FranceMember

Hi,

I've noticed that the code for the annotations ABHet and ABHom (AlleleBalance) have changed since the 3.7 version (commit 6156b85, "Fixed logic error and tidied AlleleBalance and AlleleBalanceBySample").
Before the 3.7 version, each count was weighted according to its genotype quality while since the 3.7 version, counts are no more weighted so that basically the ABHet ABHom ratios are just count ratios.
I know this annotation is "experimental" but I wonder why these changes have been made for ?

Thanks,

Best Answer

Answers

  • SheilaSheila Broad InstituteMember, Broadie, Moderator admin

    @bgrenier
    Hi,

    I need to ask the developer and get back to you.

    -Sheila

  • SheilaSheila Broad InstituteMember, Broadie, Moderator admin

    @bgrenier
    Hi again,

    The developer says:

    Weighting by genotype quality was an arbitrary and meaningless way to do this. For example, two het samples could have GQs of 50 and 100, in which case both are overwhelmingly likely to truly be hets. It makes no sense to weight the GQ=100 variant twice as much. If you're going to do any weighting, it shouldn't be by GQ, which is the negative log 10 of the probability that the genotype is wrong. Rather, it should be by the probability that the genotype is right, which for these two samples is 1 - 10^(-5) and 1 - 10^(-10), respectively, that is, essentially identical to 1 in both cases.

    I hope this helps.

    -Sheila

  • bgrenierbgrenier FranceMember
    edited August 2018

    @Sheila ,
    Hi,

    In GATK 3.6 version and before, each count was indeed weighted by 1-10^(-GQ/10) (so the probability that the genotype is right) and not directly by GQ. So this do not explain why this has been modified since the 3.7 version.

    Best,

  • bgrenierbgrenier FranceMember
    edited August 2018

    @Sheila
    Hi,

    Ok, I found the old way of weighting quite easy to understand but thank you for the explanations.

    Best,

Sign In or Register to comment.