I have a problem in understanding the exact meaning of FDR. I know it is the number of false positives divided by the total number of regulated genes. But I don't understand the meaning of adjusting FDR in reporting the result of microarray.
Let's assume that you perform a single t-test (no microarray or multiple comparisons) and get a p-value of 0.05. What's the false-discovery rate of that? 5%, the same as the p-value. Suppose instead that you perform a bunch of test (e.g., when dealing with a microarray). The likelihood of observing a significant p-value increases as you do more tests. Suppose that there are actually no significantly regulated genes (I.e., the treatment had no effect). In that case you expect ~5% of the p-values to be 0.05 or less, 10% 0.1 or less, and so on. Modern FDR adjustment techniques take advantage of things like this. Since our background expectation is a uniform distribution of p-values, we can look for an enrichment of small p-values and adjust their values according to how heavily enriched things are. That's essentially how BH FDR adjustment works, others work somewhat differently. The resulting adjusted p-values can then be meaningfully interpreted. One thing to note is that the typical threshold is often a bit relaxes for adjusted values (0.1 vs. 0.05).
I have tried many times to validate microarray or RNA-seq data by qPCR. It seems to me FDR does not make any sense, while P value indeed makes sense. if you set FDR at 0.1 or 0.05, you will almost loose a number of genes which actually have differential expression that could be validated by qPCR.
Dear Devon, Thanks a lot for your useful explanation about the exact meaning of FDR. Regards Nazanin