Hi,

I have the following experiment: two cell types (Treg/CD4), 3 donors (52/64/55), 2 batches (24/26), 3 conditions (ha/ut/cd19). My main two questions are the following: 1) what genes are DE in the two cell types (for each condition separately, so Treg/ha vs CD4/ha and so on); 2) What genes are DE in ha condition vs ut/cd19 (for each cell type separately, for example Treg/ha vs (Treg/ut and Treg/cd19).

When I project my samples with MDS, that's what I see:

So there is a clear separation between samples, but I am confused, which term I should include in my limma model. For example, donor 64 (and batch 26 at the same time) is clearly separated by cell type (CD4 or Treg). At the same time, donors 52 and 55 are also separated, even though they have the same condition, batch, and cell type. Should I include both donor and batch terms?

Thank you

Forget batch effect, it looks confounded with donor.

So you have 3 donors for each cell-type/treatment combo, one of those donors is much different from the other two.

You likely don't have the power to see the differences you want to see. Differences between donors swamps the differences between treatments and celltypes. I expect your p-values will be rotten, and I doubt there are any magic math tricks you can do to fix this.