Reference map of human cancer-associated fibroblasts (CAFs) for ProjecTILs annotation

This CAF map was generated by extracting fibroblast and fibroblast-related populations from a pan-cancer single-cell fibroblast atlas (Gao et al., Cancer Cell, 2024; DOI: 10.1016/j.ccell.2024.07.007). The source atlas compiles previously published and in-house single-cell RNA-seq datasets, comprising 517 samples across 11 tissues (breast, colon, skin, lung, kidney, pancreas, liver, ovary, stomach, head and neck, and synovium) covering both healthy and pathological states. From this resource, we manually selected a subset of samples covering these 11 tissue types to ensure broad biological representation and to approximate a balanced number of fibroblasts across tissues and conditions. Cells were processed in Seurat, computing QC metrics (UMI counts, detected genes, mitochondrial and ribosomal content) without automatic filtering. We leveraged the original study’s fibroblast cluster annotations as an initial reference and refined them through iterative semi-supervised integration using STACAS combined with a curated blocklist of genes. After the first integration, newly generated clusters were evaluated for homogeneity (>75% purity threshold). Inconsistent or contaminant clusters, as well as proliferating cells, were removed. The final cleaned dataset was downsampled to ensure balanced representation across fibroblast states and converted into a ProjecTILs-compatible reference map.