Dear all,

I would like to request your help and opinions on a meta-analysis setup utilizing two completely independent cancer cohorts, where one acts as discovery data set, while the other is meant as validation cohort.
To 'integrate' both cohorts, I would like to select genes by FDR in cohort 1 (q < 0.05) and then only test selected genes in cohort 2 (to reduce the number of tests).

Question: do I still need to do a p-value correction for the results of my validation? I would assume that I need to, as I still perform multiple tests in the validation cohort (Fisher test for each gene). However, the null hypothesis of 'no difference between cancer and normal' was already rejected before, and my expectation would be that most of the tested genes are in fact different, so the resulting distribution of p-values is non-uniform and skewed towards 0. Can FDR handle this situation and am I not throwing out too many true positives to control the number of false positives in this case? Can this be solved by lowering the threshold for FDR in the validation cohort?

Any recommendations and comments are greatly appreciated!