EQTL FDR vs raw p-value

Question

[Filago]

Hello,

I am using with GWAS data and use the FUMA-online annotation tool for this task. Amongst other things, I am especially interested in mapping eqtls from GTEx to my variants. Here comes my question: FUMA reports for each eqtl the FDR and the p-value.

How would you in general evaluate an eqtl-association: By FDR or by the raw p-value? For some variants the FDR is lower than the p-value and for others the other way around (does this mean sth.?)

I think that FDR makes more sense --> My strategy would be to filter for eqtls with a FDR < 0.05 Does this make sense to you?

Best,

Andreas

[German.M.Demidov]

As for me - absolutely. In general, you should always prefer FDR over raw p-values (given that you do this type of analysis - frequentist multiple testing).

[Carlo Yague]

For some variants the FDR is lower than the p-value and for others the other way around (does this mean sth.?)

Isn't that impossible ? As far as I know, the FDR-corrected pvalue is, by definition, always higher (or equal in very specific cases) than the uncorrected pvalues.

[Filago]

I rechecked it and it is the q-value and actually not the FDR, which is reported by this tool. Sorry for this! So in some cases the q-value is lower than the p-value. However, still I don't know if this makes sense and if yes, if it makes sense to filter for q < 0.05?

[German.M.Demidov]

It would be very strange for me to call some value q-value in the context of statistics and denote something other than FDR estimation.

[German.M.Demidov]

Possible =)

https://stats.stackexchange.com/questions/92918/q-value-less-than-p-value