Skip to content

pdac_data_sources.md

Latest commit

 

History

History
195 lines (122 loc) · 4.23 KB

pdac_data_sources.md

File metadata and controls

195 lines (122 loc) · 4.23 KB
Raw

PDAC Validation Data Sources

This file separates simulation outputs from real validation data.

What We Have Now

The files in results/ are synthetic model outputs. They are useful for:

  • checking equations
  • comparing therapy schedules
  • debugging the workflow
  • generating expected qualitative behavior

They are not real biological validation.

Tier 1: First Validation Datasets

TCGA-PAAD / GDC

Link: https://portal.gdc.cancer.gov/projects/TCGA-PAAD

Use for:

  • KRAS, TP53, SMAD4, CDKN2A mutation status
  • RNA-seq expression
  • clinical survival
  • tumor stage and grade
  • first-pass PDAC cohort validation

Primary validation question:

Do rogue-closure features derived from driver genes, immune genes, stromal genes,
and glyco-shield genes predict survival or aggressive phenotype better than
KRAS status alone?

cBioPortal TCGA PanCancer Atlas PAAD

Link: https://www.cbioportal.org/study/summary?id=paad_tcga_pan_can_atlas_2018

Use for:

  • fast exploratory mutation/expression/survival analysis
  • downloadable clinical tables
  • comparing model-derived feature sets against standard oncogenic features

GEO GSE71729

Link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE71729

Use for:

  • primary PDAC
  • metastatic PDAC
  • normal samples
  • tumor/stroma subtype validation

Key validation question:

Does the model's stroma/shield axis separate activated stroma, normal stroma,
primary tumor, and metastatic disease?

GEO GSE62452

Link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62452

Use for:

  • 69 pancreatic tumors
  • 61 adjacent non-tumor samples
  • survival-associated expression validation

GEO GSE28735

Link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE28735

Use for:

  • 45 paired tumor/non-tumor PDAC samples
  • early differential-expression validation
  • survival-linked gene screen replication

Tier 2: Perturbation And Drug-Response Validation

DepMap / Cancer Dependency Map

Link: https://depmap.org/portal/download/

Use for:

  • PDAC cell-line dependencies
  • KRAS, EGFR, STAT3, glycosylation, stromal-proxy vulnerabilities
  • CRISPR/RNAi dependency support
  • drug sensitivity context through linked DepMap resources

Validation question:

Are model-selected intervention targets dependency-supported in PDAC-lineage
systems?

NCI Patient-Derived Models Repository

Link: https://pdmr.cancer.gov/

Use for:

  • PDX / organoid / cell-line model discovery
  • PDAC model selection for future perturbation tests
  • bridge from patient omics to model systems

PDX Finder

Link: https://www.pdxfinder.org/

Use for:

  • finding PDAC PDX models
  • checking available molecular and treatment metadata
  • external model-system validation

Tier 3: Glyco-Shield Evidence

GlyGen

Link: https://www.glygen.org/

Use for:

  • glycoprotein annotation
  • glycosylation sites
  • building the initial glyco-shield gene/protein panel

Candidate glyco-shield panel:

MUC1, MUC4, MUC16, ST6GAL1, ST3GAL1, ST3GAL4, FUT3, FUT8,
B3GNT3, B4GALT1, MGAT5, GALNT3, GALNT6, SDC1, SDC4, GPC1,
HAS2, LGALS1, LGALS3

GlyConnect

Link: https://glyconnect.expasy.org/

Use for:

  • glycoprotein and glycan relationship lookup
  • manual curation of glyco-shield features

Tier 4: Imaging And Spatial Validation

The Cancer Imaging Archive

Link: https://www.cancerimagingarchive.net/

Use for:

  • PDAC imaging cohorts when available
  • tumor/stroma/radiomics proxy validation
  • future spatial closure validation

Validation Milestones

  1. Build real-data feature table from TCGA-PAAD:
sample_id, survival, stage, KRAS, TP53, SMAD4, CDKN2A,
driver_score, immune_score, stroma_score, glyco_score, rogue_closure_score

  1. Test whether rogue_closure_score predicts survival or aggressive phenotype.

  2. Repeat in GSE71729 for primary/metastatic/stroma subtype separation.

  3. Use DepMap to check whether proposed intervention axes are dependency-supported.

  4. Publish synthetic output separately from real-data validation results.

Repository Rule

Do not commit large raw datasets directly to GitHub. Commit:

  • fetch scripts
  • small manifests
  • checksums
  • derived feature tables when license permits
  • validation reports

Use releases, Zenodo, DVC, or external object storage for larger artifacts.