Fig. 2

The performance of stDyer on 12 slices from the DLPFC dataset generated by 10x Visium technology. a The histology image of slice 151673 of the DLPFC dataset. b The visualization of spatial domain annotation of slice 151673. c Spatial domain clustering and ARI scores of different methods on slice 151673. d The Silhouette scores of the 6 methods that can generate unit embeddings in the latent space. e UMAP plots of unit embeddings generated by the 6 methods in d. f The histology image of slice 151674 of the DLPFC dataset. g The visualization of spatial domain annotation of slice 151674. h Spatial domain clustering and ARI scores of different methods on slice 151674. i The Silhouette scores of the 6 methods that can generate unit embeddings in the latent space. j UMAP plots of unit embeddings generated by the 6 methods in i. k The boxplot of ARI scores of different methods on all 12 slices. The p-values were computed by the Wilcoxon rank-sum test (****: p-value < 0.0001; ***: p-value < 0.001; **: p-value < 0.01; *: p-value < 0.05). l Visualization of SVG expression values (bottom) and their associated spatial domains (top). ARI: adjusted rand index. stDyer (KNN): stDyer with a KNN spatial graph