Fig. 7

The performance of stDyer on 13 slices of E16.5 mouse embryos from Stereo-Seq. a Visualization of spatial domains predicted by stDyer, SpaGCN, BayesSpace, CellCharter, and SpaDo on slice 7 (the largest slice). b Annotated spatial domains on slice 7. c Predicted spatial domains by stDyer on slice 7. The same color would be used for each mapping between an annotated domain and matched predicted domain(s). d Visualization of SVGs for the corresponding spatial domains in c. For each SVG, the gene expression values that are higher than the median values across all units are adjusted to the maximum values for better visualization. e Heatmaps of ARI scores of 10 spatial domain clustering tools across 13 slices. The slice IDs are sorted in ascending order based on the unit numbers. The gray cells indicate methods fail to run on the corresponding slices. For each slice, we highlighted the maximum ARI in bold and the minimum ARI in italic. f Host CPU memory usage of SpaGCN. g Graphics memory usage of stDyer. h The runtime of five methods on datasets of different sizes. i The impact of batch size on runtime of stDyer to analyze 100,000 units. j The impact of batch size on GPU memory usage of stDyer to analyze 100,000 units. k The impact of the unit number on runtime of stDyer. l The impact of GPU number on runtime of stDyer to analyze 1,000,000 units