Hi all,

I referred to the discussion about camera and roast, and used camera to conduct competitive tests on Reactome data sets, and then found some interesting sets. Now I want to do mroast tests on these small interesting sets.

However, what confuses me is whether I should distinguish these sets and perform roast tests separately? For example, 5 interesting sets contribute to mitochondria function, and 3 interesting sets contribute to apoptosis. Should I mix these 8 interesting sets for roast, or test the mitochondria and apoptosis-related sets separately?

Depending on the number of input sets, the p-value and FDR value generated by mroast will be different. Although these differences do not affect the final results for now, I am still worried that if I continue to expand the size of detection sets or use different test data, it will make my final results difficult to interpret.

All the best!

The p-values from mroast will be the same (up to sampling variability associated with random rotations) regardless of whether you test all the sets together or separately. You can use fry() or increase the number of rotations to decrease the sampling variability and make the p-values more reproducible.

However, it seems that you used camera to choose these interesting sets, so the gene sets have been cherry-picked, so it not meaningful to apply FDR adjustment to these interesting sets, no matter how you do it.