The sequencing of the transcriptomes of single-cells, or single-cell RNA-sequencing, has now become the dominant technology for the identification of novel cell types and for the study of stochastic gene expression. In recent years, various tools for analyzing single-cell RNA-sequencing data have been proposed, many of them with the purpose of performing differentially expression analysis. In this work, we compare four different tools for single-cell RNA-sequencing differential expression, together with two popular methods originally developed for the analysis of bulk RNA-sequencing data, but largely applied to single-cell data. We discuss results obtained on two real and one synthetic dataset, along with considerations about the perspectives of single-cell differential expression analysis. In particular, we explore the methods performance in four different scenarios, mimicking different unimodal or bimodal distributions of the data, as characteristic of single-cell transcriptomics. We observed marked differences between the selected methods in terms of precision and recall, the number of detected differentially expressed genes and the overall performance. Globally, the results obtained in our study suggest that is difficult to identify a best performing tool and that efforts are needed to improve the methodologies for single-cell RNA-sequencing data analysis and gain better accuracy of results.
Keywords: assessment; benchmark; differential distributions; differential expression; single-cell RNA-seq.