etest calculation 'control' groups and p-value adjustment

[ManuelMoradiellos]

Hi! Thanks for creating this tool.

I wanted to ask about the behaviour of the 'control' parameter on etest() and how it affects the latter multiple testing correction.

As far as I understand from the source code of that function, when you provide a list of perturbed genes as control groups to test again it does the multiple testing correction within a loop per each of the groups and then it aggregates the results, not applying the correction to the whole set of comparisons.

Wouldn't that lead to inflated adj. p-values and wouldn't it be better to do the adjustment for all aggregated interactions at the end?

I'm testing some ideas with the Replogle dataset and after a lengthy etest computation, done for 10k permutations, the vast majority of my results were significant for my criteria (< FDR10%) but I noticed that the non-adjusted p-values seem to be all significant and after correction they keep being so, when that is not a usual behaviour for some correction methods.

Again, thanks for the help and sorry for the lengthy message

[tessadgreen]

Apologies that we haven't had the chance to investigate this—at the moment, the re-implementation at pertpy is being more actively maintained, so I'd recommend trying that instead.