Please tell me if I can use FDR twice for multiple correction in a two-stage study using omics data. For example, based on RNA-seq data, I would like to identify genes that differ in gene expression between 20 patients and 20 healthy subjects. I am considering a two-stage research design. First, the Student's t-test will be performed on the 10,000 genes as a discovery stage (edgeR or other analysis methods can be used, but since this question is about multiple comparisons, I will use the t-test). The obtained P-values are corrected for false discovery rate (FDR), and q-value < 0.05 is considered significant. As a result, 50 genes showed significant differences. As the next stage, I validate these 50 genes in samples different from those measured by RNA-seq. Another 20 patients and 20 healthy subjects were measured by methods other than RNA-seq and compared again by t-test. 50 times were tested, so I have to do multiple correction. Can I adapt the FDR again here and assume that q-value < 0.05 is significant? Or should I apply the correction to Bonferro at this stage and set the significance level at P value < 0.05/50 (=0.001)? I would be grateful if you could enlighten me.
Please, use edgeR and not t-tests. It is beyond me why people still think this gives any advantage. What does it mean "50 times were tested"?
Thank you for your prompt comment. I am sorry in that context, but I am not asking what statistical analysis method should be used for group comparisons. The t-test is just an example. I just want to know if it is possible to use FDR as a correction method for multiple comparisons in both the discovery and validation stages. As for the meaning of “50 times were tested”, since 50 genes were significant in the discovery stage, 50 tests will be performed to validate these 50 genes in the validation stage. Can I also apply the FDR to the correction for multiple comparisons in the validation stage? I wanted to ask.