VQSR / CNN filtering for small (~100) gene panels

Hello,

I'm trying to perform germline variant calling on a panel with ~100 genes. I was wondering what the bare minimum (in terms of sample size) would be for variant filtering via VQSR. If the sample size is prohibitively high given the size of the gene panel, would it be appropriate to pad the training set with exome data from 1000G or ExAC? In general, is there a minimum number of variants needed to train the model?

Also, is there a timeframe for the official release of GATK CNN, and would something like this be applicable to my gene panel once available?

Thanks for your help!
Will

Best Answer

  • SheilaSheila Broad InstituteMember, Broadie, Moderator
    edited March 14 Accepted Answer

    @willhooper
    Hi Will,

    While we don't have exact numbers of variants that are appropriate for VQSR, we recommend using at least 30 exome samples or a few whole genome samples. Your dataset may be too small for VQSR, so you may be better off trying hard filtering for now.

    The CNN should be available to test out very soon and hopefully be better suited for your dataset :smile:

    -Sheila

Answers

  • SheilaSheila Broad InstituteMember, Broadie, Moderator
    edited March 14 Accepted Answer

    @willhooper
    Hi Will,

    While we don't have exact numbers of variants that are appropriate for VQSR, we recommend using at least 30 exome samples or a few whole genome samples. Your dataset may be too small for VQSR, so you may be better off trying hard filtering for now.

    The CNN should be available to test out very soon and hopefully be better suited for your dataset :smile:

    -Sheila

Sign In or Register to comment.