Comparative analysis of single-cell RNA-seq cluster methods

The emerging Single-cell transcriptome sequencing technologies give rise to new resource for cell biology. Transcriptomic landscapes of heterogenetic samples at the single-cell resolution enable characterization of cell sub-types and reveal gene co-expression pattern. Numerous efficient algorithms have been developed to accurately normalize, cluster and visualize cells from single-cell transcriptome sequencing profiles, including but not limited to Seurat, SC3, SIMLR, and SCANPY. However, systematic comparisons of the performance of these scRNA-seq cluster method are lacking. Here, we use 7 gold-standard scRNA-seq datasets with clear label and Tabula Muris, a dataset of millions of single-cell transcriptomes, to evaluate the 4 scRNA-seq cluster method. Results shows that SCANPY is more time-cost-efficient for large-scale data but SC3 is more precise for cell sub-types recall. Our quantitative comparison offers an informed choice among 4 scRNA-seq cluster methods, and it provides a hint for further improvements of scRNA-seq analysis methods.