Extended Data Fig. 10: Robustness analysis of Scriabin’s binning workflow.

a-d) Mouse and human PBMC scRNA-seq datasets from 10X Genomics were analyzed. a) UMAP projections of mouse and human PBMCs colored by the sample of origin (left) and by manually-annotated cell types (right). b) Heatmap depicting overlap between bin identity and cell type annotations. Each row sums to 100%, and the annotations at left show the number of cells within each bin and maximum degree of overlap of each bin with a given cell type identity (ie. the highest value in each row). c) UMAP projection highlighting cells in bin #191. d) Bar plot depicting differentially-expressed genes in bin #191 relative to other B cells shared between the human and mouse cells in bin #191. Differential expression tests were run individually for human and mouse cells. e-g) A toy dataset of ~14,000 peripheral blood mononuclear cells (PBMCs) with nine sub-datasets was analyzed. e) Density plot depicting the number of cells in each bin. The median bin size in this analysis is 25 cells. f) As in (b) An SNN graph was used to assess cell-cell connectivity for the binning workflow. Cell type annotations are transferred from a reference dataset and are thus orthogonal to the data used to generate the bins. g) Dot plot depicting the cell type annotations and scores for the anchor pairs used to generate the bins depicted in (f).