Do Benjamini-Hochberg p.adj values get less strict when analysing only one superfamily only? RNA-seq

Question

[foejvs546]

Hi everyone!

Do Benjamini-Hochberg p.adj values get less strict when analysing only one superfamily instead of the whole transcriptome? I would expect there to be fewer false positives, is that correct?

Also, is this FDR-correction approach valid, since I will only be analysing one superfamily instead of all genes?

I hope this isn't horribly ignorant. Thank you.

Thank you!

Answer 1

[Istvan Albert]

The BH method works by looking at the distribution of the p-values, it does not factor in "what" you are analyzing.

In the simplest terms, the BH method assumes that when there is no effect the p-values have a uniform distribution. If you were to have an effect that would distort that uniform distribution.

Another way to think of it is that it evaluates how many significant p-values were produced by the background noise alone.

But again, note how adjusting p-values is not about what category you choose or what you analyze.

It is about figuring out a correction based on the distribution of the p-values.

[foejvs546]

Hi Istvan, thank you very much. That's really helpful.