edgeR multiple testing

[colindaven]

Hi everybody,

I've searched a lot but have not found a definitive discussion on this yet.

I'm using edgeR on a human dataset with 3-4 replicates per group.

When I follow the edgeR tutorial and export (see code below) I get this:

dispersion = estimateCommonDisp(DGE_timepoints)
de.com <- exactTest(dispersion, c("healthy-control","disease"))
out_tags1 = topTags(de.com, n=length(de.com))
write.table(out_tags1, "out_unadjusted_pvalue.csv", sep="\t")

gene table.logConc table.logFC table.PValue table.FDR adjust.method
ENSG00000250644 -20.0951079864 -11.2384935084 1.19615633839674e-08 0.0002112173 BH
ENSG00000176994 -33.1995219248 -33.633064638 1.27917856074067e-07 0.0009680458 BH

Note that adjust method is BH, but I am not sure if Pvalue adjustment was carried out by default. In the manual it seems to suggest a further Pvalue adjustment is needed:

# P values adjusted by Benjani-Hochberg method, single column
out_tags_adjusted = p.adjust(de.com$table$p.value, method="BH")

write.table(out_tags_adjusted, "out_adjusted_pvalue.csv", sep="\t")

However, the two top results above then get an "adjusted" Pvalue of 1 and 0.3 respectively, and virtually no hits are significantly differently regulated.

I guess that the initial step is sufficient to implement multiple testing.
Comments?

[Dario1984]

You don't need to do any more adjustment. The adjusted values are in the column which is called 'table.FDR'.

[colindaven]

Ah I see. Thanks for the help!