Pre-computed TCGA-BRCA downstream analysis results for:

DELPHI: Trustworthy Spatial Molecular Profiling from Histopathology Images

Enables reproduction of Figure 5 (survival analysis) without downloading ~2 TB of TCGA whole-slide images.

Extract into results/tcga_brca/at the repository root.

Contents

• spatial_features.csv — 17 niche-derived spatial features for ~1000 patients
• univariate_cox_by_pam50.csv — PAM50-stratified univariate Cox regression
• multivariate_cox.json— 3-feature Cox model (C-index = 0.7546, P = 2.66e-12)
• niche_annotations.csv — Per-patient niche tissue-type labels
• annotations— Per-patient niche annotations (~1000 JSON)
• niches_expression— Expression-based niche assignments (~1000 NPZ)
• niches_confidence— Confidence-based niche assignments (~1000 NPZ)
• inference — Representative DELPHI inference outputs (9 PT files)
• ablation— Confidence-filter ablation variants 

Usage

tar -xzf delphi_tcga_results.tar.gz -C results/tcga_brca/
python scripts/figures/plot_fig5_main.py

Output: figures/main/fig5.png

Provenance

Generated by:
scripts/survival/infer_brca_full.py
scripts/survival/run_pipeline.py --once
scripts/figures/plot_fig5_main.py