SCOPE-DEL

Similarity & COverage Prioritization and Evaluation for DNA-Encoded Libraries (SCOPE-DEL): Notebooks, data, and results for GTM-driven chemography to prioritize DNA-Encoded Libraries (DELs) by balancing similarity and chemical-space coverage.

---

Table of contents

• Repository structure
• Quick start (PDM)
• Data
• Reproducing figures & tables
• Results
• Contributing
• License
• Acknowledgments

---

Repository structure

SCOPE-DEL/
├── notebooks/
│   └── Analysis_results.ipynb            # Metrics, comparisons, figure export
├── data/
│   ├── raw/                              # Input sources (unmodified)
│   └── processed/                        # Cleaned data ready for modeling
├── results/
│   ├── dels_100/                         # Metrics
├── docs/
│   └── assets/
│       └── SCOPE-del-logo.png            
├── pyproject.toml                        # PDM project config
├── pdm.lock                              # Locked dependency graph (generated)
├── LICENSE                               # MIT
└── README.md

---

Quick start (PDM)

Requires Python 3.10+ and PDM 2. If you don’t have PDM:

pipx install pdm   # recommended
# or: pip install -U pdm

1. Create and activate the PDM virtualenv

# inside the repo root
pdm venv create -i python3.10     # or your preferred interpreter
pdm use -f .venv/bin/python        # on Windows: .venv\Scripts\python.exe

2. Install dependencies

pdm install

Notes:

• RDKit wheels are platform-dependent. The project uses rdkit/rdkit-pypi. If installation fails on your OS, see RDKit’s wheel/conda guidance or switch the fingerprinting backend in the notebooks.

3. Run the notebooks

# (optional) register a Jupyter kernel named “SCOPE-del”
pdm run python -m ipykernel install --user --name SCOPE-del

# start Jupyter
pdm run jupyter lab     # or: pdm run jupyter notebook

4. Recreate exports

• Run notebooks/DataPreprocessing.ipynb to populate data/processed/.
• Run notebooks/GTM_GMM_optimization_benchmark.ipynb for GTM map selection.
• Run notebooks/Analysis_results.ipynb to produce figures/tables in results/.

Handy commands

# freeze dependencies for external runners/CI
pdm export -f requirements -o requirements.txt --without-hashes
# run quality tools if configured
pdm run black . && pdm run isort .

---

Data

• data/raw/ contains references (or instructions) for obtaining large sources not committed to the repo.
• data/processed/ is generated by DataPreprocessing.ipynb (fingerprints, deduped sets, splits).
• For reproducibility, the preprocessing notebook documents exact retrieval, cleaning, and descriptor parameters.

---

Reproducing figures & tables

Analysis_results.ipynb exports:

• Correlation heatmaps comparing GTM-space vs fingerprint-space metrics
• Coverage-vs-similarity scatter plots
• Density/class landscapes
• Summary CSVs with overlaps, EF@k, and other selection metrics

Artifacts are written to results/figures/ and results/tables/ with filenames matching manuscript labels.

---

Results

Key takeaways reproduced by the notebooks:

• GTM-derived, centroid-oriented metrics serve as practical proxies for pairwise fingerprint measures in DEL selection.
• Visual GTM landscapes help balance similarity to a reference with broader chemical-space coverage.

---

Contributing

Issues and PRs are welcome. Please:

• Keep notebook outputs deterministic where possible.
• Note OS/Python/PDM versions when reporting issues.
• Discuss large additions (new DEL panels, alternative metrics, new GTM maps) in an issue first.

---

License

This project is licensed under the MIT License. See LICENSE for details.

---

Contributors

• Alexey Orlov aorlov@unistra.fr - contributor to the development of ChemographyKit; performed machine learning, interpreted the data, and contributed to manuscript writing.
• Dragos Horvath dhorvath@unistra.fr - provided overarching guidance, conceived and planned the research, and supervised the overall project.
• Alexandre Varnek varnek@unistra.fr - provided overarching guidance, conceived and planned the research, and supervised the overall project.
• Louis Plyer louis.plyer@unistra.fr - contributor to the development of ChemographyKit; performed machine learning, interpreted the data, and contributed to manuscript writing.
• Tagir Akhmetshin tagirshin@gmail.com - contributor to the development of ChemographyKit.
• Erik Yeghyan varnek@unistra.fr - preparation of ChEMBL datasets used in this study.
• Fanny Bonachera varnek@unistra.fr - preparation of ChEMBL datasets used in this study.

---