Inquiry regarding FDR adjustment with covariates

I am writing to inquire about the default FDR correction method in MaAsLin2.

It appears that the BH adjustment is applied to the entire results table by default. For instance, with 100 taxa and 3 fixed effects (e.g., treatment, age, and sex), the correction is performed over 300 tests rather than 100.

Could you please clarify the rationale for including covariates (e.g., age and sex) in the total test count for the main variable of interest (e.g., treatment)? Is it considered statistically acceptable to extract the p-values for the main variable and re-calculate q-values based solely on the number of features (e.g., 100), as in the previous post (e.g., Maaslin2 handling of covariates, FDR adjustment in MaAslin2-update)? I would appreciate your guidance on the recommended practice for such cases.

Thank you for your time and for this excellent tool.

Hi,

FDR correction should always be performed over all the p-values that might be of interest. If you were not interested in any significant age or sex associations and were only interested in treatment associations, you could FDR correct over only the treatment p-values. However, if you’re interested in associations with age, sex, or treatment, you should correct over them all. We correct over everything by default to be conservative.

Will