jeffbhasin/methylaction

Question about covariate in regressions

Opened this issue · 1 comments

The covariate is used only in Stage 2's testDESeqANODEV function.

However, Stage 1's negative binomial testDESeq (used for pairwise pre-tests) and Stage 2's negative binomial testDESeq (used for pairwise post tests) does not use the covariate.

Should Stage 2's post test control for covariates?
Is there a reason why the covariates are not controlled in these NB tests?

Hi Ming,
I think you are correct about stage 2 post test not using covariates, because it is using the exact test and not the GLM to my recollection in the post-test. We were comfortable that if the ANODEV GLM was significant (that did use the adjustment), we were not worried about the covariate in the post test. However, I think it could be adapted in the code to use the GLM there instead in the stage 2 post-test if you wanted to be sure to incorporate the covariate.

Jeff