EdgeR with Paired Sample Replicates

Question

[kmyers2]

I am running EdgeR to find differentially expressed genes and I want to make sure I am constructing the model and contrast correctly.

I have 9 total samples, 3 experimental samples (ExpA, ExpB, ExpC) and 6 controls samples (CtrlA1, CtrlA2, CtrlB1, CtrlB2, CtrlC1, CtrlC2). The samples with the same letter (ExpA, CtrlA1, CtrlA2 for example) were collected on the same day - thus the collections occurred on three different days. I need to find the genes that are differentially expressed between the experimental samples and the paired control samples but want to take into account the paired nature of the samples.

I have set up a group in EdgeR to use Exp/Ctrl and Day as grouping factors:

group <- factor(paste(targets$Experiment, targets$Day, sep = "."))

That yields six levels: Ctrl.A, Ctrl.B, Ctrl.C, Exp.A, Exp.B, Exp.C

I then use this to construct the design:

design <- model.matrix(~0+group)

That yields a design matrix that I use to perform the estimations and fit. Then to set up the differential comparison test, I use the following contrast:

my.contrasts <- makeContrasts(DipvsCtrl = (Exp.A-Ctrl.A)+(Exp.B-Ctrl.B)+(Exp.C-Ctrl.C), levels = design)

Quickly comparing the counts data seems to suggest this is the correct contrast to make, but I just wanted to see if anyone else could notice something I missed. Did I set up the contrast correctly to make the comparison I want to make?

Thanks!

Answer 1

[Benn]

If you want to include "paired" in your design, you should not make the group factor, but include both experiment and day in the design.

experiment <- factor(targets$Experiment)
day <- factor(targets$Day)
design <- model.matrix(~0+experiment+day)

# Depending on the column names of your design:
my.contrasts <- makeContrasts(ExpvsCtrl = Exp - Ctrl, levels = design)

[kmyers2]

Thanks! So if I understand, this will maintain the days in the comparison right? It is best to compare samples collected on the same day (ExpA vs CtrlA). I just want to make sure I'm understanding it correctly.

[Benn]

Yes you are testing "exp" vs "ctrl" with this design, but with the "day" effect taken into account (think of it as a similar approach to a paired t-test).

[kmyers2]

Thank you very much! That makes sense to me and I really appreciate the help!