PDAC Tissue Explorer – processed data: tissue proteogenomics, single-cell atlas summaries, clinical data and PDAC model systems

Processed data behind the PDAC Tissue Explorer, an interactive web application for exploring where human genes are expressed in pancreatic ductal adenocarcinoma (PDAC) tissue: tumour versus matched adjacent normal tissue, tumour versus healthy pancreas, cell types, tissue states (healthy donor, adjacent normal, treatment-naive primary tumour, liver metastasis, treated primary tumour), and published fibroblast and tumour-cell subtypes.

The files are derived from three public resources and are provided so that the application can be run locally and the figures it produces can be reproduced. They contain summarised or already openly released values only; no raw sequencing data or individual-level controlled-access data are included.

Version 0.3.0 adds the app's on-disk data store, CPTAC phosphosites, clinical data (survival, stage, classical/basal-like score of the CPTAC and TCGA-PAAD tumours), single-cell co-expression partners and PDAC model systems (DepMap cell lines and organoids, GDC ORGANOID-PANCREATIC and HCMI organoids).

Files. Data store read by the app (version 0.3.0; SQLite, read with the R packages DBI and RSQLite, one compressed record per gene plus small sample tables; decoded by app/R/01_data.R of the code repository):

• genes.rds: searchable gene index; gene_xrefs.rds: Ensembl identifiers and gene types (links to other resources)
• cptac.sqlite: CPTAC PDAC protein and RNA, tumour and paired adjacent normal
• bulk.sqlite: TCGA-PAAD and GTEx pancreas RNA (recount3)
• sc.sqlite: per-sample cell-type summaries and atlas map of the single-cell atlas
• phospho.sqlite: CPTAC site-level phosphoproteome, tumour and paired normal
• clinical.sqlite: survival, stage and classical/basal-like score of the CPTAC and TCGA-PAAD tumours
• coexp.sqlite: most co-expressed genes within each cell type across treatment-naive primary tumours
• models.sqlite: DepMap Public 26Q3 RNA of all cancer models, CRISPR knockout effects (Chronos) and proteomics; RNA of PDAC organoids (GDC ORGANOID-PANCREATIC, HCMI) and of the HCMI organoids' source tumours; model subtype and resemblance scores Tables from version 0.1.0 (R serialised objects, readRDS() in R >= 4.3; Matrix needed for sc_umap.rds), unchanged: cptac.rds, bulk_tcga_gtex.rds, sc_groups.rds, sc_expr.rds, sc_umap.rds, genes.rds. data-raw/05_build_store.R builds cptac.sqlite, bulk.sqlite and sc.sqlite from them. Download everything checksum-verified with get_data.R from the code repository.

Single-cell processing. Source: the integrated PDAC single-cell atlas of Loveless et al. (Clin Cancer Res 2025; Zenodo 10.5281/zenodo.14199536), which combines 12 published studies; tumour-cell and fibroblast subcluster labels come from the authors’ repository. Before summarisation, the following were removed: the cycling-cell cluster; cells with high ambient-RNA contamination (decontX score > 0.30, or above the 98th percentile within each sample and cell type); and cells with fewer than 200 detected genes. This left 618,841 cells from 234 samples. Expression values are the atlas’s log-normalised values (log1p of counts per 10,000). Ductal cells are split into the atlas’s nine tumour subclusters and the remaining ductal cells; fibroblasts into myofibroblast (myFb 1–5) and inflammatory (iFb 1–3) subtypes.

Reproducibility. All files were produced by the scripts in the source-code repository (data-raw/), which also contains provenance logs with input checksums, cell, gene and sample counts, and R session information. Additional scripts: 01b (co-expression), 02b (phosphosites), 03b (clinical), 04b (identifiers), 06a/06 (model systems); provenance logs with input checksums and counts are in data-raw/logs.

Notes and limitations. TCGA and GTEx are separate studies, and no cross-study batch correction was applied. Adjacent normal tissue from cancer surgery is not healthy pancreas. In the single-cell atlas, tissue state and treatment are partly confounded with the contributing study. Clinical associations are exploratory (small cohorts, short follow-up). Model systems are compared only within one technology (cell lines with cell lines, organoids within their collection); their resemblance to patients is measured against tumour cells of the single-cell atlas.

Please cite this dataset together with the original data sources: Cao et al., Cell 2021; Wilks et al., Genome Biol 2021 (recount3); The Cancer Genome Atlas Research Network, Cancer Cell 2017; The GTEx Consortium, Science 2020; Loveless et al., Clin Cancer Res 2025, and the single-cell studies integrated in that atlas. DepMap, Broad (2026), DepMap Public 26Q3 (CC BY 4.0) and Tsherniak et al., Cell 2017; Nusinow et al., Cell 2020; Tiriac et al., Cancer Discov 2018; ElHarouni et al., Nature 2026 (HCMI); Liu et al., Cell 2018 (TCGA-CDR); Moffitt et al., Nat Genet 2015. The models and the data are from the Human Cancer Models Initiative (HCMI); dbGaP accession number phs001486.