Hi,

I've got a paired (or blocks really) experiment with the setup like the below.

Five blocks and four conditions (A, B, C, D). I want to contrast all vs all (A vs B, A vs C, A vs D and etc.). The Xs in the table below indicate a sample present for each block (I think it's clearer in table format rather than colData format). Clearly there are a lot of "missing pairs", but there are at least two paired samples when comparing any of the conditions.  

So, my question is what would be the best way to do an analysis with DESeq2?

As far as I can see, it's either keep all the data and use the design ~Block + condition then extract each contrast from the results object, or subset the data  and keep the design balanced. It's noted that block 3 and 5 (-B5) cover half the contrasts.

Reading previous forum posts the consensus seems to be subset the data or use limma-voom, but the posts I've read don't deal with multiple conditions.  

Greg

The case where I recommend that users use duplicateCorrelation() is a little different. That is where it is desired to estimate condition effects for different groups of subjects, but a subset of the subjects only have one level of condition,. Here there is not the same nesting of subjects within biologically-relevant groups. Here you want to estimate condition effects, which are assumed to be the same regardless of block, while controlling for block effects. You should be able to just use ~block + condition.