Fig. 5: SIDISH identifies malignant high-risk populations and prognostic biomarkers in PDAC 10x Xenium spatial transcriptomics dataset.

a Spatial atlas of annotated cell types across the pancreatic tumor section, including fibroblasts, B cells, CXCL9/10 cells, T cells, mast cells, lymphatic endothelial cells, macrophages, endothelial cells, endocrine cells, tumor cells, metaplastic cells, smooth muscle cells, acinar cells, and ductal cells. b UMAP visualization showing strong concordance between high-risk cells (41,323) and tumor cells, with high-risk tumor cells (27,887) in red, high-risk non-tumor cells in cyan, and background cells in gray. c Bar plot showing the proportion of high-risk cells by cell type, with tumor cells representing 67.5% of all high-risk cells, followed by acinar cells (15.7%) and metaplastic cells (7.7%), indicating strong tumor enrichment. d Confusion matrix quantifying overlap between high-risk and tumor cells, revealing a 74.5% concordance (McNemar’s test P = 3.54 × 10⁻¹⁴⁶). P values were calculated using McNemar’s test to evaluate the association between tumor and high-risk cell labels. e Functional enrichment bar plots. The SIDISH-derived marker genes are enriched in biological processes and pathways related to PDAC progression and poor survival. f Kaplan–Meier survival curves for the independent bulk validation dataset (GSE224564). High-risk patients, shown in pink, exhibit clear stratification from background patients, shown in gray (P = 2.42 × 10⁻¹¹), validating the clinical relevance of SIDISH-derived marker genes. P values were calculated using the two-tailed log-rank-sum test to compare survival curves between high-risk and background patient groups.