Hello,

I am interested in testing for differentially expressed genes with edgeR across multiple groups, using the ANOVA-like approach.

For example, in the mouse mammary gland experiment in the edgeR user's guide, contrasts are defined and top differentially expressed genes obtained as follows:

con <- makeContrasts(
            L.PvsL = L.pregnant - L.lactate,
            L.VvsL = L.virgin - L.lactate,
            L.VvsP = L.virgin - L.pregnant, levels = design
)
anov <- glmQLFTest(fit, contrast = con)
topTags(anov)

How should one best approach the question of which of these genes are significantly different for the individual comparisons?

If instead the samples had been processed using voom from the limma package, the decideTests function has two methods, "global" and "nestedF", which could be appropriate for this purpose; however, decideTests in edgeR does not have a similar argument.

One can use glmQLFTest repeatedly, selecting the individual comparisons:

qlf.L.PvsL <- glmQLFTest(fit, contrast = con[, "L.PvsL"])
qlf.L.VvsL <- glmQLFTest(fit, contrast = con[, "L.VvsL"])
qlf.L.VvsP <- glmQLFTest(fit, contrast = con[, "L.VvsP"])

Should one then take the unadjusted p-values from these three tests and then adjust them all together, something like the "global" approach?

Thanks!

Hi Jamie,

The edgeR and limma authors usually test each of the contrasts separately, which is the default in both limma and edgeR. However, if you wanted to use the "global" method in edgeR, you could do exactly what you propose. Just cbind the p-values together and apply decideTests:

P <- cbind(PvsL = qlf.L.PvsL$table$PValue,
           VvsL = qlf.L.VvsL$table$PValue,
           VvsP = qlf.L.VvsP$table$PValue)
Results <- decideTests(P, method="global")

The choice between separate or global multiple testing adjustments depends on how you want to report the results. It doesn't make any difference whether you did the anova test to start with -- the contrast tests are in either case separate to the anova tests.