Supplementary Figure 1: Comparison of annotations of PBMC data with reference-based methods for cell type annotation.

This figure accompanies Fig. 1c. t-SNE plots of the PBMC-4K scRNA-seq data²¹. The plots are colored by the top annotation using the methods presented by Kang et al.²² (top left), Li et al.²³ (top right) and SingleR (bottom left). Kang et al. correlated 173 differentially expressed genes learned by comparison of scRNA-seq clusters of PBMCs. Li et al. (RCA) used a bulk microarray reference dataset for correlation. SingleR introduced a fine-tuning step to refine correlation with bulk datasets. While all methods agreed on the general annotations of monocytes, T cells, B cells and NK cells, annotations differed for cellular subtypes: RCA annotated the left-most cells in cluster 4 as NK cells, while the method of Kang et al. and SingleR annotated them as CD8⁺ T cells, which was supported by the expression of individual genes CD3E (general T cell marker) and CD8A. It is of note that NKG7 and GNLY are commonly used as NK cell markers but are also expressed in activated CD8⁺ T cells. The method of Kang et al. and RCA annotated the cells in cluster 3 as CD4⁺ T cells, while SingleR annotated 47.7% of this cluster as naive CD8⁺ T cells and 30.4% as central memory CD8⁺ T cells (T_CM). The CD8A marker, along with CCR7 and SELL (markers for naïve cells), supported this annotation. The expression of IL7R (marker for memory T cells) in some of those cells suggested that this cluster also contained memory cells, as annotated by SingleR. FOXP3 (a marker of regulatory T cells) was found only in a small proportion of T_reg cells (in the sorted T_reg cells, only 5.6% of cells expressed FOXP3); however, SingleR suggested that many of the cells in cluster 1 were in fact T_reg cells. In this comparison, only SingleR was able to distinguish highly similar cell states, and these differential annotations were supported by gene markers viewed individually.