Relationship between FDR alpha level and adjusted p-value thresholds

Question

[basch]

I've been running a gene expression experiment, using DESeq2 for differential expression analysis, but I have problems understanding FDR using the BH method.

Suppose I use an alpha of 0.1 (default in DESeq2), what happens if I use a significance threshold of 0.05 for the adjusted p-values? Is this possible or do the alpha (0.1) and significance threshold have to be the same to make sense?

Despite watching videos and reading tutorials, I have trouble understanding how to interpret the adjusted p-values and how they relate to the alpha

Answer 1

[Istvan Albert]

The interpretation of FDR is not the same as the interpretation of a "traditional" p-value.

The FDR makes sense only in a sorted list of p-values and refers to the proportion of false discoveries up to that point (everything above).

So if in row 200 you have an FDR of 0.05 with 200 rejected null hypotheses up to that point (aka differentially expressed transcripts) it means that 200*0.05 = 10 of those rejections are likely to be invalid.

So you can see that the alpha will have a different meaning when applied to a p-value or an FDR.

FDR adjusted p-values should be called q-values and that would avoid confusion, but then of course people would complain that the tool does not generate p-values ...

[basch]

Thank you for the explanation! I think I got some better understanding of this concept. But, even when using FDR, DESeq2 has a parameter called "alpha" in the results function (the default value is 0.1). This has something to do with independent filtering, but I don't understand how the choice of alpha affects the adjusted p-values (or q-values).

[Istvan Albert]

There is an explanation on alpha here:

https://support.bioconductor.org/p/104618/

from reading it appears that it is a convenience parameter to trim the list to a given cutoff.