Hi all,

I have a general question about the model estimation used by DESeq.  There have been many posts on whether or not DESeq works well with a small number/no replicates, but I'm wondering if it's appropriate to use DESeq for differential analysis across two conditions, where one condition has a small number of replicates (say, 5) and the other has a huge number (in the 100s).  The particular phenotype that we are looking at in our (clinical) data is quite rare, but we'd still like to test for differential expression.  Usually I would provide a reproducible example but these data are sensitive...

In this case, does it make sense to use DESeq?  Would it make sense to, say, randomly sample some of the replicates from the condition with 100s of replicates and run multiple tests?  I'm reading the original DESeq2 paper to try to understand how the model is built but any tips would be much appreciated.  Thank you!

You shouldn't downsample the condition with 100s of replicates. There is nothing special to do, but you should note that the dispersion estimate will be mostly influenced from the group with more replicates. You can look at plotCounts afterward to verify that the top genes make sense, and are not affected by any artifact.