Extended Data Fig. 9: Human universal fibroblasts.

a, Top, UMAP embeddings of human pancreatic cancer tumour and normal adjacent tissue (n = 21,626 cells). Bottom, percentage of cells in each cluster coming from tumour or NAT. b, UMAP as in a, coloured by expression of indicated genes. c, Relative average expression of top 10 marker genes (sorted by log(fold change)) for each cluster in the pancreatic cancer single-cell dataset. Two representative genes highlighted per cluster. DEGs across clusters. d, Top, expression level of indicated marker genes (colour, y-axis) across 100 pseudo-bulk samples (x-axis) generated from human pancreatic cancer scRNA-seq data. The known percentage of cells from cluster 8 in each pseudo-bulk is shown by the dotted blue line. Bottom, boxplots representing the distributions of pairwise correlation coefficients of the top 20 marker genes for cluster 8 in pseudo-bulk samples containing (left) and not containing cells from cluster 8 (right). e, Boxplots summarizing DPT expression distributions across tissues from the GTEx portal. Tissues with mean above horizontal black line were included in correlation analysis (f). n = 7,851 biologically independent samples. f, Co-expression as in d, results from the gene-by-gene correlation matrices are summarized as boxplots for each individual tissue from GTEx. n = 5,957 biologically independent samples. g, Gene-by-gene correlation matrix of pairwise correlations in 205 normal pancreas bulk RNA-seq samples from GTEx. Blue indicates Pi16⁺ cluster signature gene, red indicates Col15a1⁺ signature gene. h, Human universal fibroblast score projected onto human pancreatic cancer samples. i, Human universal fibroblast score projected onto human subcutaneous adipose. d–f, Box and whisker plots show minimum and maximum (whiskers), interquartile range (box) and median (centre line).