Straightforward clustering of single-cell RNA-Seq data with t-SNE and DBSCAN

Clustering of cells by cell type is arguably the most common and repetitive task encountered during the analysis of single-cell RNA-Seq data. However, as popular clustering methods operate largely independently of visualization techniques, the fine-tuning of clustering parameters can be unintuitive and time-consuming. Here, I propose Galapagos, a simple and effective clustering workflow based on t-SNE and DBSCAN that does not require a gene selection step. In practice, Galapagos only involves the fine-tuning of two parameters, which is straightforward, as clustering is performed directly on the t-SNE visualization results. Using peripheral blood mononuclear cells as a model tissue, I validate the effectiveness of Galapagos in different ways. First, I show that Galapagos generates clusters corresponding to all main cell types present. Then, I demonstrate that the t-SNE results are robust to parameter choices and initialization points. Next, I employ a simulation approach to show that clustering with Galapagos is accurate and robust to the high levels of technical noise present. Finally, to demonstrate Galapagos accuracy on real data, I compare clustering results to true cell type identities established using CITE-Seq data. In this context, I also provide an example of the primary limitation of Galapagos, namely the difficulty to resolve related cell types in cases where t-SNE fails to clearly separate the cells. Galapagos helps to make clustering scRNA-Seq data more intuitive and reproducible, and can be implemented in most programming languages with only a few lines of code.