DESeq(2): multi-conditions assay

[fabou]

Dear SEQanswer community

I face the following question and wonder how to approach it with DESeq or DESeq2:

I have four conditions (each with replica) A, B, C, D whereas A and B be are the controls and C and D where treated with similar but different agents. Now I'd like to identify the genes which are more pronounced differential regulated due to D than due to C.

How can I approach this? Is it advised to do it ...

• ... pairwise and consider only genes which are significant in B vs. D but not in A vs. C (which ommits genes which are induced by both agents but with different strength)
• ... with condition<-c(untreated, untreated, untreated, treated) and no Type and no GLM
• ... with condition<-c(untreated, untreated, treated, treated) and Type<-('S1','S2','S1','S2') with GLM modelling count ~ Type + condition

As you might figured from the question I do not fully get what the GLM does.

Any help is highly appreciated
thank you very much in advance

fabou

[mastal]

Why do you have two sets of controls, what is the difference between A and B?

[Michael Love]

hi Fabou,

To get at the right approach, i would also need to know what's the difference between A and B.

But if A and B are identical controls, you can compare the log2 fold change from treatment2 over treatment1 using a contrast.

If you define a condition variable with levels "control","treatment1","treatment2", and a design:

design(dds) <- ~ condition

Then you can test for genes with differential expression comparing treatment 2 vs treatment 1 using:

res2vs1 <- results(dds, contrast=c("condition","treatment2","treatment1"))

You could also combine multiple results, for example genes with low adjusted p-value from this comparison, as well as from a contrast of treatment 2 vs control, in order to define set of genes which are differentially expressed in treatment2 vs control, and vs treatment1

res2vsC <- results(dds, contrast=c("condition","treatment2","control"))

res2vs1$both <- (res2vs1$padj < 0.1 & res2vsC$padj < 0.1)

[fabou]

Thank you Michael, I will explore the strategy you outlined.