Fig. 5

Using vissE to identify and characterise a cancer associated fibroblast (CAF) phenotype in spatial transcriptomics data of a breast cancer patient. a A Haematoxylin and Eosin (H&E) image of the breast cancer tissue profiled using the 10X visium technology. b Spots profiled coloured by the projection of the first principal component (PC1). c Pathologist’s annotations of stromal (olive-green), malignant (purple) and mesenchymal-like (gold) regions of the tissue overlayed on the H&E image. d A gene-set overlap graph of gene-sets enriched in PC1 with nodes representing individual gene-sets and edges representing overlaps based on the adjusted rand index (ARI). Nodes are coloured based on the direction and significance of enrichment: green nodes represent gene-sets enriched in PC1-high spots. Six gene-set clusters representing biological themes are identified, containing 69, 21, 8, 6, 11 and 5 gene-sets respectively. e Cluster annotations generated by text-mining analysis of gene-set names. f Gene loadings (also known as weights) for genes belonging to gene-sets in the cluster plot against the number of gene-sets in the cluster the gene belongs to. g Protein–protein interaction (PPI) networks between genes that belong to gene-sets in the cluster. Each node represents a gene and edges represent known PPIs. Nodes are coloured based on gene loadings (also known as weight). h Cell type deconvolution (left) and expression of CAF-related marker genes (center) for the top 20% of spots with the highest PC1 projection vs. all other spots (region marked in the right panel)