Fig. 10

Alignment of pancreatic islet cells captured using three different protocols identifies cell type-specific variation across protocols. a–d UMAP visualization of pancreatic islet cells sequenced on CEL-Seq, CEL-Seq2, and Smart-Seq2 after alignment by scAlign, colored by protocol, cell type, clustering on the alignment space, or scAlign’s state variance map. e Scatterplot indicating the overlap of clusters defined using the state variance map (y-axis) and based on the cell type labels as reported in Stuart et al. f Comparison of clusters identified using the embeddings, versus using the state variance map. Shown are two clusters defined in the embedding space, termed alpha-1 and alpha-2 because of their overlap with the alpha cell type. Gray points in the alpha-1 plot indicate cluster 2 cells, and gray points in the alpha-2 plot indicate cluster 1 cells. Colored points represent the three clusters identified in the state variance map. scAlign’s variance map clusters (1, 2, and 3) are each found in both alpha-1 and alpha-2, indicating poor agreement. g Heatmap of the state variance map computed across the three capture protocols (CeL-Seq, CEL-Seq2, and Smart-Seq2) where red indicates high variance of expression predicted for a given gene and cell across protocols