Fig. 4

geneBasis delineates cell types regardless of their abundance and resolves unannotated inter-cell-type variability. Panels A–F correspond to mouse embryogenesis; G, H correspond to spleen, and I, J correspond to pancreas analysis. A Cell type confusion matrix for the first 100 selected genes (prog. = progenitors; mes. = mesoderm; Brain = Forebrain/Midbrain/Hindbrain). B Ordered bar plot representing relative cell type abundance. C Upper panel: UMAP representation of Cardiomyocytes, colored by clusters from Tyser et al, when integration between datasets was performed using all genes. Lower panel: UMAP representation of Cardiomyocytes, colored by mapped cardiac clusters, when integration between datasets was performed using selection from geneBasis. UMAP coordinates themselves were calculated using the whole transcriptome. D Confusion matrix for cardiac clusters, where the mapping using all genes is used to assign true identity and mapping using the geneBasis selection is used to represent a mapped cluster. E Upper panel: UMAP representation of Gut tube, colored by mapped organs (Nowotschin et al.), when integration between datasets was performed using all genes. Lower panel: UMAP representation of the Gut tube, colored by mapped organs, when integration between datasets was performed using the selection from geneBasis. UMAP coordinates themselves were calculated using the whole transcriptome. F Confusion matrix for the emerging organs present along the gut tube, where mapping using all genes are used to assign cells to a true cluster and the geneBasis selection is used to represent a mapped cluster. G Cell type confusion matrix generated using the first 100 selected genes. For cell types in spleen, we use abbreviations, see Table S2 for full annotations. H Ordered bar plot representing relative cell type abundance. I Cell type confusion matrix for the first 100 selected genes. J Ordered bar plot representing relative cell type abundance