Fig. 3

Establishing cell type marker genes in snRNA-seq reference data. A Schematic of an ideal deconvolution cell type marker gene heatmap, where the marker genes for a target cell type (rows) only have high expression in the target cell type (columns), and low expression for all other cell types. B Illustration of marker gene selection strategies for an example target cell type, Oligo. (i) 1vALL combines the non-target cell types into one group and identifies differentially expressed genes between the two groups. (ii) Mean Ratio maintains all the cell type groups, finds the ratio between the mean expression of the target cell type, and the highest mean expression from a non-target cell type. C Scatter plots of the Mean Ratio value vs. 1vALL standard log fold change, for all genes by cell type. The top 25 ranked by Mean Ratio are indicated by point color, and the top gene from both methods is annotated with the gene symbol. D Violin plots of the gene expression over cell types for the top gene from 1vALL (top row) and Mean Ratio (bottom row) methods. E Heat map of the top 5 marker genes from 1vALL logFC. Rows are genes, columns are pseudobulked samples by cell type and tissue block. Color is the Z-score of logcount expression scaled by gene. F Heat map (similar to E) of the top 5 Mean Ratio marker genes