An ideal testing procedure in the analysis of RNA-seq datasets should be able to incorporate the inherent uncertainty of abundance estimates, as well as all nuisance variabilities. Most popular methods for RNA-seq differential expression analysis fit a parametric model to the counts or scaled counts of reads for each gene or transcript. If the underlying parametric model does not fit the data well, these methods may have an inflated number of false discoveries. Previous work has shown that non-parametric algorithms for RNA-seq differential expression may in some cases have better control of the false discovery rate. While some parametric models have been proposed which incorporate the uncertainty of abundance estimates into the differential expression testing, existing non-parametric methods do not take into account the inferential uncertainty of the observations, which we show may also lead to an inflated false discovery rate. We propose a non-parametric testing procedure applied to inferential replicate datasets, which both accounts for nuisance covariates such as library batch and also for the inferential uncertainty of the abundance estimates, that better controls the FDR.