Fig. 3

Smoother enhances cell-type deconvolution performance in various spatial omics data. a–c Detection of tumor T-cell infiltration in a 10x Visium ductal carcinoma sample [8]. a Left to right: Images showing single channel CD3 immunofluorescence staining (FITC/green), T-cell proportions as estimated by Smoother-guided NNLS and CARD, respectively. b Spearman correlation of predicted T-cell proportions with CD3 fluorescence levels across all spots, with whiskers indicating the 95% confidence intervals. c Spearman correlation of predicted T-cell proportions with the expression of T-cell marker genes across all spots, similar to b. d-g Spatial mapping of neurons in the 10x Visium mouse brain section ST8059048 [27]. d Left: H&E staining. Right: UMAP visualization of the paired single-nucleus transcriptomic reference of 52 neural subtypes. e Brain subregions identified by clustering based on estimated cell-type compositions. f-g Estimated relative abundances of cortical (f) and hippocampal (g) excitatory neurons. Color intensities are proportional to the estimated cell-type proportion and are scaled by the same factor across methods for visualization purposes. HPC, hippocampus; DG, dentate gyrus (CA4). h–k Spatial mapping of embryonic cell types in the spatial-CUT&Tag H3K4me3 50-µm data of mouse embryo [4]. h Left: Overview of the spatial-CUT&Tag technology, adapted from [4]. Right: T-SNE visualization of the scRNA-seq reference of 37 embryonic cell types [37]. i Embryonic subregions identified by clustering based on estimated cell-type compositions. j Estimated relative abundances of five major lineages aggregated from individual cell types. Color intensities are proportional to the estimated cell-type proportion and are scaled by the same factor across methods for visualization purposes. k Standardized gene activity of selected marker genes