I am testing many (500,000) genetic variants, and the tests are FDR corrected and give me a q-value. Normally I would just call everything with q < .05 significant. But in this case I am testing those same genetic variants in two other related experiments (not using exactly the same individuals, but the samples may overlap). What to do? Would changing the significance threshold for q to .05/3=.0167 be an option? Many thanks!
The answer would depend on how you measure errors (and their proportions). If you are concerned with the proportion of false discoveries within each experiment, then do separate FDR corrections. If you are worried about the "global" proportion of false discoveries, you could treat all the experiments as one. This would guarantee Global FDR control, but the FDR within each experiment is NOT controlled for. I believe your suggestion is a conservative way to get FDR control at both levels: within each experiment and globally.