Single-cell RNA-sequencing (scRNA-seq) has quickly become an empowering technology to characterize the transcriptomes of individual cells. Although many early analyses of differential expression (DE) have focused on finding markers for cell sub-populations (experimental units are cells), there is now an emergence of datasets across replicates and multiple conditions where the goal is to make patient-level inferences (experimental units are patients), with 100s to 1000s of cells measured for each patient. This indeed provides an opportunity to go back and make use of the existing robust bulk RNA-seq frameworks, by first aggregating the data into "pseudobulk" counts at the subpopulation level. However, this opens up many new questions, which we will address in this talk: how does one track subpopulations across patients (e.g., in the presence of batch effects)? do we lose information by aggregating (i.e., is it better to model the single cell data directly)? if using pseudobulks, how do we do normalization? We will present a comprehensive framework for flexible multi-sample multi-condition DE of scRNA-seq experiments.