Good day!
I'm trying to carry out co-expression analysis using CEMiTool after limma preparation of microarray results.
It's pointed in CEMiTool userguide that one should use unprocessed expression data.
Experimentally I found that there is no big difference between evaluated co-expression modules if I change FDR p.value in topTable function from 0.05 to 0.1.
But it appears that there really is a big difference whether to use topTable(...adjust.method = "BH" or "none" before submitting data to CEMiTool - the genes changes their positions in co-expression modules.
Should I use the Benjamini-Hochberg correction? Or maybe I should not filter data by p.values and correct it at all?
The values I use are average Lfc's, each from 3 repeatings.
What do you mean by: Experimentally I found that there is no big difference between evaluated co-expression modules if I change FDR p.value in topTable function from 0.05 to 0.1? selecting only some genes?
If you select the option "none" in your toptable, you are carrying all false positives and false negatives which are around 5%. If you are comparing 10.000, you are dealing with ~ 500 genes ( false +/-).
Best
Thank you for your reply!
I meant that the particular modules of gene co-expression remains mostly the same - genes doesn't make any drift between them. In this case the new modules are added, but my aim is to study the co-expression module containing one particular gene of interest, so it is not crucial. Anyway, after the modules evaluation I extract the ID's of genes belonged to one particular module, combine it with expression levels and p.values and filter away everything that have adj.p>0.05 and |lfc| < 1.
Yep! that´s make sense to me. Adjusted p-value does not have a big impact in such case but neither hurt. In my opinion, applying BH and selecting p-values above 0.05 as you did should be fine.
Best
- Badges
- Science topic
- Similar topics
- Correction