Circulating cell populations as response predictors and targets to improve immunotherapy in metastatic lung cancer

Repository containing two complementary analyses related to immunotherapy response in lung cancer:

• ML_analysis/: classical machine learning pipeline (Random Forest + SHAP) across clinical, cytokine, metabolite, and cell-type features.
• scRNAseq_analysis/: single-cell RNA-seq workflow focused on T cell populations from GSE179994.

Repository layout

lung_cancer_responses_ML/
├── ML_analysis/
│   ├── environment.yml
│   ├── README.md
│   ├── data/
│   ├── int/
│   ├── results/
│   └── src/
└── scRNAseq_analysis/
    ├── README.md
    └── Scripts/

1) ML analysis (Python)

What it does

Builds and interprets Random Forest classifiers to predict responders vs non-responders from multiple feature sets:

• clinical
• cytokines
• metabolites
• cells
• all features combined

Environment

From ML_analysis/:

conda env create -f environment.yml
conda activate lung_cancer_ml

Typical run order

From ML_analysis/:

1. Clean and integrate exported TSVs

python src/clean_data.py

2. Train RF + generate SHAP outputs (set dataset inside script)

python src/rf_shap_analysis.py

3. Evaluate all top-feature combinations

python src/feature_combinations.py

4. Build R heatmap visualizations

Rscript src/plots.R

Outputs are written to ML_analysis/results/.

For details, see ML_analysis/README.md.

2) scRNA-seq analysis (R)

What it does

Runs preprocessing, clustering, and post-treatment CD4/CD8 subsetting for GSE179994 T cells, supporting responder vs non-responder downstream analyses.

Requirements

• R 4.3.0+
• Core packages: Seurat, dplyr, scuttle, ggplot2
• Optional: SeuratDisk (for .h5ad export)

Main entrypoint

From scRNAseq_analysis/:

Rscript Scripts/run_analysis.R

Outputs are written to scRNAseq_analysis/Figures/ and scRNAseq_analysis/Results/.

For full input/output and methods details, see scRNAseq_analysis/README.md.

Data notes

• The ML pipeline expects intermediate TSV files under ML_analysis/int/.
• In the ML workflow, key raw TSVs were manually exported from source spreadsheets before running scripts.

Suggested citation

Manuscript in preparation.