Annotating single-cell RNAseq clusters by similarity to reference single-cell datasets

Single cell RNAseq is often used to examine cell types within tissue samples. There are a multitude of methods available for clustering sequenced cells into transcriptionally-similar groups, putatively corresponding to cell type or state. However, once the clusters are defined it can be difficult to determine (a) what cell type each cluster might represent and (b) how well the clustering method has reconstructed the cell-groups at a level relevant to the biological question of interest. This work aims to be able to take pre-computed cell-clusters and annotate possible cell type information to each cluster in a quick, accessible manner on the basis of similarity to publically available single-cell datasets. This will be done by comparing genes differentially expressed in a specific cluster of an experiment (compared to the rest of the experiment), with equivalent pre-computed signatures from publically available reference datasets. Initial experiments show a promising recapitulation of biologist-annotated cell types between public human and mouse brain tissue datasets. Further work will determine how well such an approach might identify shared cell types (e.g. endothelial cells) across different tissue types, and if this technique could be used to evaluate the selection of a particular set of cell cluster definitions in an experiment, on the basis of their biological relevance.