Fig. 3: HEARTSVG identifies tumor related SVGs and predicts spatial functional domains with distinct biological functions in the 10X Visium colorectal cancer data.

a Original hematoxylin and eosin stained (H&E) tissue image (left) and results of unsupervised spatial clustering (right). The red-circled areas in the HE image represent the tumor regions. b The bubble plot illustrates the results of KEGG pathway enrichment analysis for 19 tumor-related pathways (x-axis) across different methods. Each bubble represents a pathway, and its size corresponds to the overlap gene size of the pathway and SVGs detected by each method. The x-axis and y-axis of the plot represent different methods and their significance (\({-\log }_{10}({{{\rm{p}}}}-{{{\rm{value}}}})\)). P-values were calculated by g:Profiler. c The ROC curves illustrate the TPR and FPR of six different methods when using common gene modules of tumor microenvironments (left) and consensus molecular markers of colorectal cancer subtypes (right) and as gold standards for true SVGs. AUC is the area under the ROC curve. d HEARTSVG predicts four spatial domains based on SVGs and graphed the average expression of SVGs in each spatial domain. e Representative SVGs correspond to the four predicted spatial domains in Fig. 3d. f Enrichment analysis corresponds to the four predicted spatial domains. The length of bars represents the enrichment using \({-\log }_{10}({{{\rm{p}}}}-{{{\rm{value}}}})\). P-values were calculated by g:Profiler. Source data are provided with this paper.