CNS_ML

Analysis code and processed data for the manuscript From Genomic Instability to Prognosis: Copy Number Signature Clusters as Predictive Biomarkers Across Tumors.

The repository contains the analysis notebooks, processed inputs, and figure outputs used in the paper. Raw TCGA data and source files from the three copy number signature studies are not redistributed.

Overview of the study workflow (Figure 1a of the manuscript).

Repository structure

CNS_ML/
├── README.md
├── LICENSE
├── LICENSE-DATA.md
├── environment.yml
├── assets/
│   └── figure1_panel_a.png
├── data/
│   ├── README.md
│   └── processed/
│       ├── *.RData
│       └── zenodo/
├── code/
│   ├── 00_data_preprocessing.Rmd
│   ├── 01_signature_exploration.Rmd
│   ├── 02_clustering.Rmd
│   ├── 03_survival_analysis.Rmd
│   ├── 04_machine_learning.Rmd
│   └── 05_methylation_batch_effect.Rmd
└── Methylation_batch_ correction_result/

Reproducing the analyses

Create the R/Python environment with:

conda env create -f environment.yml
conda activate cns_ml

The notebooks should be run in numerical order. Notebooks 01, 02, 03, and the Supplementary Figure 6 section of 05 can be run using the processed files included in this repository. Notebook 00 documents the complete, computationally intensive preprocessing workflow and uses external input paths.

Notebook 04 records the model specifications, hyperparameter grids, and performance-extraction procedures used in the study. Serialized trained models are not distributed because some were created with incompatible historical package versions. They can be retrained from final_matrices.RData using the documented stratified split and model code.

Analysis notebooks

Script	Manuscript content
code/00_data_preprocessing.Rmd	Multi-omics preprocessing and stratified train/test splitting
code/01_signature_exploration.Rmd	Signature distributions, PCA, Pearson correlations, Jaccard analyses
code/02_clustering.Rmd	Patient clustering and cluster composition
code/03_survival_analysis.Rmd	Kaplan-Meier curves, stratified Cox models, survival tables
code/04_machine_learning.Rmd	Continuous signature prediction and cluster prediction models
code/05_methylation_batch_effect.Rmd	Methylation platform harmonization, batch-effect checks, and Supplementary Figure 6

Data

Small processed inputs required by the notebooks are stored in data/processed/. The large starting matrices are distributed through the associated Zenodo record:

• final_matrices.RData contains the final Drews, Steele, and Tao multi-omics matrices. These matrices allow users to recreate the train/test partitions and retrain any of the models described in 04_machine_learning.Rmd.
• exp_meth_post_normalization.RData contains ordinata, the harmonized pan-cancer expression and methylation matrix used by the preprocessing workflow.

Place the downloaded files in data/processed/zenodo/. File contents and dimensions are listed in data/processed/zenodo/README.md.

The folder Methylation_batch_ correction_result/ contains the per-tumor diagnostic plots used to identify the Infinium I/II probe-design bias and assess BMIQ normalization.

Outputs

The figures can be generated by running the corresponding .Rmd notebooks. The final publication-quality figure files are provided with the manuscript rather than duplicated in this repository.

Citation

The manuscript citation and Zenodo DOI will be added after publication of the corresponding records.

License

Code and analysis notebooks are licensed under the MIT License. Original documentation, figures, and processed data produced for this project are licensed under CC BY 4.0.

Source data and third-party materials, including material derived from TCGA, the NCI Genomic Data Commons, and the cited signature studies, remain subject to their original access conditions, licenses, and citation requirements.