This is a copy of this post https://github.com/drisso/zinbwave/issues/58#issue-656953981 but I thought it might reach out to more people by posting it here too.

Hi everyone,

I am measuring differential expression between two conditions in 10X data, using zinbwave and DESeq2. I am using K=2 in zinbwave as there are some latent factors (e.g. library size) that I would like to infer from the data. Now, would it make sense to include W1 (as calculated with zinbwave) in the DESeq2 design model (something like design <- model.matrix(~condition + reducedDim(exp, "zinbwave")[,1])? Or is W somehow already reflected in the observational weights, in which case including it in the model would be redundant?

Similarly, as I use the LRT test in DESeq2 should I include W in the reduced formula?

Many thanks for your input! Vincent

I'm not sure but I believe that the latent factors are not redundant with the observation weights. I'll await the zinbwave team to answer definitely.

Yes, if you included a latent factor in the full design which is controlling for unwanted variation, you should also include it in the reduced design.