I am working on an RNA-seq dataset for differential expression: 4 environments, 4 biological replicates per environment. There are also 4 batches, but they are orthogonal to environments, so all environments are equally represented among all batches.
I lean towards EdgeR. I find it more user friendly and have used it more often, easier to select for a common-sense effect size in addition to p value, it has a robust way of estimating variation, uses negative binomial distribution (but this is experimental in lme4) and so on.
However, some senior colleagues wants to use the mixed model approach of lme4 so that batch will be a random factor. Why does EdgeR not allow the use of random factors? What kind of arguments could I use in favor of EdgeR over lme4? Or if that shows my bias too much, what are the strengths of each approach?