Dear All,

I have performed RNAseq on an experimental design using 6 samples subject to two rounds of treatments: The first treatment uses a molecule that induces a signalling pathway (T1), the second round of treatment uses two drugs that disrupt that signalling pathway (D2_A and D2_B). The drugs used in the second round are used independently (i.e. drug 2_A and drug 2_B are not used together in any one sequencing reaction). The samples are paired, such that each of the 6 samples was subject to all combinations of the treatments.

As such I have an experimental design that looks something like this (for the first sample), where NT=no treatment 1 and ND = no treatment 2:

6 (samples) x 2 (treatment 1) x 3 (treatment 2) = 36 libraries

I have a range of questions to ask of the data but most importantly, I want to know how each of the second treatments (the drugs) modifies the effect of the first treatment (the induction of a signalling pathway). I.e. what is the difference between T1-NT with D2A vs. T1-NT with ND.

Would I need to do something special when building the design matrix, i.e. am I looking to create a nested interaction formulae, or could I use a simpler additive matrix then handle the comparison using a range of subtractions when selecting the coefficients?

I currently have something like this:

> Group <- factor(paste(design$T1,design$T2, sep = "_"))

> levels(Group)

[1] "NT_D2B"     "NT_ND"   "NT_D2A" "T1_D2B"    "T1_ND"  "T1_D2A"

> design.mat <- model.matrix(~0+Group+SampleID)

> colnames(design.mat)

[1] "GroupNT_D2B" "GroupNT_ND" "GroupNT_D2A" "GroupT1_D2B" "GroupT1_ND" "GroupT1_D2A" "SampleID2" "SampleID3" "SampleID4" "SampleID5" "SampleID6"
 

Then make my contrasts (just drug D2A shown):

> cont.matrix <- makeContrasts(levels = design.mat,

                            standard.T1.vs.NT = GroupT1_ND-GroupNT_ND,

                            D2A.T1.vs.NT = GroupT1_D2A-GroupNT_D2A,

                            T1.diff.D2A.vs.standard = ((GroupT1_D2A-GroupNT_D2A)-(GroupT1_ND-GroupNT_ND)))

Does this look like a reasonable approach considering the biological question “how is the induced pathway affected by treatment with the drug?”

Kind regards,

Donny

Your current design matrix and contrasts seem sensible to me. The only thing I would suggest is to use the glmQLFit and glmQLFTest framework, rather than glmFit and glmLRT; see this paper for more details.