Hi all, I'm wondering how to interpret the logFC of an interaction term correctly. As an example, in the limma user guide section 9.5. (Interaction Models: 2 × 2 Factorial Designs), sub-section 9.5.2 (Analysing as for a Single Factor), a contrast matrix is generated as follows:

> cont.matrix <- makeContrasts(
+     SvsUinWT=WT.S-WT.U,
+     SvsUinMu=Mu.S-Mu.U,
+     Diff=(Mu.S-Mu.U)-(WT.S-WT.U),
+     levels=design)
> fit2 <- contrasts.fit(fit, cont.matrix)
> fit2 <- eBayes(fit2)

(abbreviations: Mu=mutated, WT=wild-type, S=stimulated, U=unstimulated)

So the interaction contrast Diff=(Mu.S-Mu.U)-(WT.S-WT.U) can be used to identify genes that respond to stimulation differently between mutant and wild-type samples, i.e. two factors (genotype and treatment) are combined.

Now suppose we use topTable for this "Diff" contrast, returning a log2-fold-change of 1.58 (= 3-fold upregulation) for a certain gene. Would that logFC value refer to genotype (3-fold upregulation in mutant vs. wt), or refer to treatment (3-fold upregulation in stimulated vs. unstimulated), or to both (upregulation in mutant-vs-wt AND stimulated-vs-unstimulated)?

It means that when you calculate (Mu.S-Mu.U)-(WT.S-WT.U), the value you get is 1.58. Which can mean any number of things. For instance the difference in the first set of parentheses could be 1.58 and 0 in the second. Or 0 in the first and -1.58 in the second. Or any set of numbers that you can plug in to that formula to get 1.58 back out.

If you only have a handful of significant genes, the best idea IMO is to just plot them. An alternative is to do a heatmap of your significant genes, which can be useful to show the genes with similar expression patterns. Or you can use kmeans to get groups of genes that have a similar pattern.