Bonferroni correction for enrichment

[virg4l]

I have a couple of lists that I'd like to test for enrichment of (1) pseudogenes, (2) operons, and (3) a couple of gene families. The phyper test has gone smoothly, but I'd like to run a correction for multiple testing, and I'm not really sure what I should be using as the "number of experiments" in each case.

So, for a phyper(131,1658,48246-1658,585,lower.tail=F) pseudogene enrichment test, I'm tempted to multiply the resulting pvalue by 48246 (the number of genomic features) and 11 (the number of "gene types" from Ensembl, one of which is pseudogenes).

For operons, I'd multiply the p-value by genomic features * 2 (one option for in-an-operon, one option for not-in-an-operon).

For the specific gene families, should I find the number of all gene families in my organism?

Do these plans sound reasonable, or am I way off base?

[dpryan]

For pseudogenes you would use 11, since that's the number of tests performed (how many features went into each test is irrelevant). Note that bonferroni corrected values tend to be rather conservative, so you can often loosen your normal threshold for calling something significant.

[virg4l]

Bonferroni Correction

Just 11 (and leave out the ~48,000 genomic features)?

[dpryan]

Are you testing each of the 48000 genomic features individually? If not you already have your answer.

[virg4l]

I'm not sure what you mean by individually.

I pulled all genes & their respective gene types out of Ensembl's BioMart, so I am using a gene type for each of the ~48,000 features as my background set.

I specifically care about the 585 genes in the set that I'm testing, though.

[dpryan]

I'm not sure what you mean by individually.

I know, my question was somewhat rhetorical. Bonferroni (and all other p-value corrections) only care about the number of tests performed, not the number of things used in each of the tests. The number of genes doesn't matter, just the number of tests performed.

[virg4l]

Thank you so much for your help!