DESeq analysis on subsets of multivariate data

[jnhutchinson]

I have a dataset with 7 conditions, 2 replicates each. I would like to do a number of paired comparisons of subsets of these conditions with DESeq and my question is this: as I am only looking to do pairwise comparisons, is it better to work with subsets of the data from the start and follow standard comparisons of two experimental conditions (i.e. calculate the sizefactors and dispersions for each subset separately) or is it better to work in a multivariate fashion, with size factors and dispersions calculated form the entire dataset.
If the latter, is it possible to still do pairwise comparisons on subsets of this data, perhaps by excluding some of the conditions from the GLM?

Cheers,
John

[Marcel]

I'm dealing with the same issue. My experience is that subsetting the dataset leads to more significant DE genes. But I'm not sure if this is the right way to go.

Cheers,
Marcel