Fig. 1

Generation of a scRNA-seq dataset with known cellular composition. a Schematic illustration of the experimental setup. Eight human cell lines were individually profiled by bulk RNA-seq and mixed in four batches containing mixtures of two or three cell lines each for scRNA-seq profiling. Correlation of the single-cell to bulk expression profiles was used for cell type assignment as described in the Methods section. b Visualization of correlations between single-cell and bulk expression profiles for each batch. The top row represents cell type assignment. Single cells were assigned to the cell type correlating most with their expression profile as described in the Methods section. Cells with z-scored correlations below 0.2 were not assigned to any cluster. Cells that correlate strongly with more than one bulk expression profile likely represent doublets and were excluded from future analyses. c Heatmap of gene expression values, clustered by their Pearson’s correlation across rows (genes) and columns (cells). The color bars indicate the cell type and the corresponding batch. Only the top 10% genes selected by NBDrop are shown