I am running an RNAseq analysis using DESeq2, with which I want to find out what is the effect of 2 continuous factors (A and B) and their interaction (A:B; is this the same as A*B?) on gene expression (bulk-RNA).
I have scaled and centered my continuous factors. Once I've performed the LRT with each of the reduced factors, I'd like to extract the coefficients obtained for specific genes.
If my design is: A + B + A:B And I run dds1 for the interaction term:
dds1 <- DESeq(dds, test="LRT", full = mm1, reduced = mm0)
Where mm1 and mm0 are:
mm1 <- model.matrix(~ scaled_A + scaled_B + scaled_A:scaled_B, colData) mm0 <- model.matrix(~ scaled_A + scaled_B, colData)
What is the log2FC giving me in my case? I read in an old post that the logFC was giving the coeficient or regression slope, but the design was simpler and Wald was used instead.
I can otherwise run
And I get an object with the gene list and these coefficient values for each gene
intercept scaled_A scaled_B scaled_A.scaled_B
The value in scaled_A.scaled_B coincides with the value in log2FC.. but it does as well when running dds2 (testing factor A as reduced) and dds3 (testing factor B as reduced). Is then possible that log2FC is not informative in the LRT case?
I started my analysis working with categorical factors and later I switched to continuous to get a more robust analysis. Is it possible that the LRT test gives completely different results when using my factors as categorical or continuous? Or is it not expected to change critically and then I might be doing something wrong?