A pan-cancer single-cell tumour atlas: processed data

Processed, annotated AnnData (.h5ad) objects supporting a pan-cancer single-cell RNA-seq atlas spanning 30 cancer types (1,010 specimens, 4,828,616 cells) assembled from CancerSCEM 2.0. These are derived data, not primary sequencing: the atlas is built from 64 published accessions (GEO, ArrayExpress, GSA/BioProject), listed per cancer type in accession_table.csv. Objects include the annotated cell-type atlas (481,738 cells; 17 cell types), the immune compartment (250,046 cells; 22 subtypes), and malignant cells (89,530 cells) with transcriptional program scores, plus CNV, trajectory, and cell–cell communication objects supporting the figures. Raw counts are retained in layers['counts'] where applicable. Processing: Scanpy 1.11 with Harmony integration on cancer type. See the associated manuscript Methods.