Hello,

I have samples from two tissues (brain/heart) with two level of infection (infection vs no infection) and I am using edgeR to detect differential expressed genes between the tissue according to the infection status. I am hesitating between different methods Here is the sample table

sample tissue infection indiv
A brain inf+ i1
B heart inf+ i1
C brain inf+ i2
D heart inf+ i2
E brain inf+ i3
F heart inf+ i3
G brain inf+ i4
H heart inf+ i4
I brain inf- i5
J heart inf- i5
K brain inf- i6
L heart inf- i6
M brain inf- i7
N heart inf- i7

I've made the "standard" approach

design <- model.matrix(~infection+tissue)

But I wonder if the following wouldn't be a better approach

group <- factor(paste0(exp_desc$tissue, ".", exp_desc$inf))

which gives me 4 levels: heart.inf+ heart.inf- brain.inf+ brain.inf- and then somehting like

con <- makeContrasts(brain_inf = brain.inf+ - brain.inf-,
                 heart_inf = heart.inf+ - heart.inf-,
                 brain_heart_inf = brain.inf+ - heart.inf-,
                 brain_heart_no_inf = brain.inf- - heart.inf-,
                 levels=design)

But I am not sure this is relevant. Do you have any thought? Thanks

In general, the second approach is completely standard. The first approach prevents you from distinguishing the infection profiles in brain vs heart, and detecting such tissue-specific effects was presumably the whole point of the experment.

However, I wonder whether there is another level to the experiment that is omitted from the sample table. Is it possible that this is actually a paired experiment, with brain and heart samples from the same individuals? If so, then this would really be a multilevel experiment, as hinted at in the title of your question.