Association test with covariate depending on the gene-set

[fabiennejh]

Hello,

I would like to include in my association test a covariate which depends on the gene, and use only the useful covariate for each gene. How can I do this quickly with variant association tools ? The option --covariate is fixed and, if I want to use the option --group_by Gene_name by instance, I can't use the same shortcut for covariates, even if the covariates have the same names as the genes.

Do you have any idea about it ? Thank you for your help Fabienne

[gaow]

@fabiennejh sorry I dont understand the motivation for this claim "test a covariate which depends on the gene, and use only the useful covariate for each gene." . Are you saying some covariates are specifically associated with rare variants in some genes and therefore you want to remove them for that gene in particular? From what I understand, covariates are usually meant to adjust for some global biases and should not be gene specific. So I'm trying to understand your motivation here.

[fabiennejh]

I would like to include a covariate which is gene specific (like the coverage of the gene in each individual for example). But the way I understand the command line is that the covariate is used for every set of variants of the option group_by (like the sex of the individuals for example). But the coverage of the gene A should only be used for ajustment when analysing gene A and not gene B.

[gaow]

Hmm it sounds like you want to use coverage as a proxy to genotyping errors -- the lower the coverage, the more rare variants you'll see in the sample due to genotyping error? I'm not sure if this is the best strategy though. Usually we do quality control for variants based on coverage and remove low coverage variants up-front. For my curiosity, is there a reference you point me to that does this gene-level adjustment?

Back to your question -- I cannot think of a way under the current framework that deals with the problem ... but I'll wait for input from others @jma7 @BoPeng .

[fabiennejh]

I have no special reference to give you, sorry. It's not a classical way of adjustment indeed, but I would be interested in testing it, and was wondering if it could be possible with the framework of varianttools, that I'm used to using.