Many methods have been developed for differential expression across cells using single cell RNA-seq data, few methods were proposed individual-level differential expression analysis. A straightforward solution is to add up the read counts across all the relevant cells (e.g., cells of a specific cell type) of each individual, and thus reduce the data to a pseudo-bulk RNA-seq data, with one number per gene and per individual. Then any existing methods for differential expression of bulk RNA-seq data can be applied. However, in scRNA-seq data, what is observed per gene and per individual is the distribution of gene expression across multiple cells. Reducing a distribution to a number certainly loses a good amount of information. Is it possible to directly test differential expression using gene expression distribution? We explored a few approaches, by estimating gene expression distribution using negative-binomial regression or denoised distribution estimates given by deep count autoencoder, quantifying divergence of distributions using Jensen-Shannon divergence or Wasserstein distance, and performing kernel-based association test or Permutational Multivariate Analysis of Variance.