ggo-imc

Spatial profiling of early-stage lung adenocarcinoma reveals patterns of immunomodulation and epithelial plasticity

Kim, Ravichandran, Yoffe et al. — Cancer Cell (2025)

This repository contains code to reproduce all main figures from the manuscript. The study uses Imaging Mass Cytometry (IMC) to profile 2.24 million cells across 122 early-stage lung adenocarcinoma specimens, identifying two progression trajectories (immune-inflamed and fibrotic) and a diagnostic blind spot in which 20.5% of fibrotic tumors are radiologically misclassified.

Citation: Kim J, Ravichandran S, Yoffe L, et al. Spatial profiling of early-stage lung adenocarcinoma reveals patterns of immunomodulation and epithelial plasticity. Cancer Cell. 2025. (citation to be updated upon publication)

---

Data

Processed and raw data are available on Zenodo (access may be restricted prior to publication):

• — Processed Data
• — Raw Data

---

Environment Setup

conda create -n ggo-imc python=3.9 -y
conda activate ggo-imc
pip install -r requirements.txt

---

Reproducing Figures

Run from the repository root using Snakemake. The pipeline auto-detects the container runtime (Docker on macOS, Apptainer on Linux/HPC).

All figures:

snakemake -d . -s Snakefile all

Individual figures:

snakemake -d . -s Snakefile figure1
snakemake -d . -s Snakefile figure2
snakemake -d . -s Snakefile figure3
snakemake -d . -s Snakefile figure4
snakemake -d . -s Snakefile figure5

Reset sentinel files:

snakemake -d . -s Snakefile clean

Local conda (no container):

conda activate ggo_imc
SC_TOOLS_RUNTIME=none snakemake -d . -s Snakefile figure1

Note: snakemake figure5 runs patient_group.py and roi_pca_plot_group.py (Python, no R required). The patient rule (scripts/asd.R) is a separate target not included in figure5 or all — it requires R and is not yet committed to this repo. The R-dependent panels of Figure 5 cannot be reproduced from this repo until asd.R is added.

---

Figure-to-Script Mapping

Figure	Content	Script(s)
Figure 1	Cohort overview, cell type heatmap (1a-d), ROI PCA archetypes (1e)	celltype_heatmap_info.py, roi_pca_plot.py
Figure 2	Immune dynamics: lymphocyte abundance across stages (2a-h), myeloid/macrophage polarization (2i-m)	celltype_differential_abundance.py, t_cell_analysis.py, myeloid_analysis.py
Figure 3	Stromal expansion, epithelial remodeling, EMP (3a-g)	celltype_differential_abundance.py, epithelial_characterization.py
Figure 4	UTAG microenvironments, TLS, tumor-stroma interface, cell-cell interactions (4a-f)	ue_analysis.py
Figure 5	Patient risk groups via hierarchical clustering, fibrotic trajectory, diagnostic gap (5a-e)	roi_pca_plot_group.py, patient_group.py, asd.R

---

Verifying Reproducibility

A checksum manifest of all pipeline outputs is provided in figures_checksums.md5. After running the pipeline:

python3 -c "
import hashlib
from pathlib import Path
exts = {'.pdf', '.png', '.svg'}
for line in open('figures_checksums.md5'):
    md5, path = line.strip().split(' ', 1)
    p = Path(path)
    if p.exists():
        actual = hashlib.md5(p.read_bytes()).hexdigest()
        status = 'OK' if actual == md5 else 'CHANGED'
        if status != 'OK':
            print(f'{status}: {path}')
"

No output means all figures are identical to the verified run.

---

Repository Structure

scripts/
  # Pipeline — Figures 1-5
  celltype_heatmap_info.py           # Figure 1: cell type heatmap
  roi_pca_plot.py                    # Figure 1: ROI PCA archetypes
  celltype_differential_abundance.py # Figures 2, 3: immune/stromal density
  t_cell_analysis.py                 # Figure 2: lymphocyte functional states
  myeloid_analysis.py                # Figure 2: myeloid/macrophage polarization
  epithelial_characterization.py     # Figure 3: epithelial phenotypes, EMP
  ue_analysis.py                     # Figure 4: UTAG microenvironments, TLS
  roi_pca_plot_group.py              # Figure 5: patient group PCA overlay
  patient_group.py                   # Figure 5: patient risk stratification
  # Utilities
  download_yaml.py                   # Download data from Zenodo/Box
  concat_anndata.py                  # Concatenate per-sample AnnData
  label_metadata_anndata.py          # Attach clinical metadata
  generate_manifests.py              # Generate batch manifests
  # Benchmarks
  benchmark_integration.py           # Integration method comparison
  benchmark_segmentation.py          # Segmentation benchmark
  # Archived
  exploratory/                       # Exploratory scripts (not in pipeline)