ligand-classification

ligand-classification is a configuration-driven scaffold for ligand activity classification experiments. It focuses on reproducible feature generation, consistent training pipelines, and saved artifacts you can audit later.

Quick start

conda env create -f environment.yml
conda activate lig-cls
python -m src.cli.process_features data/example/raw/psychlight_a.csv \
    --output-dir data/example/processed/psychlight_a
python -m src.cli.train --config configs/train.yaml \
    --datasets psychlight_a_descriptors \
    --models logisticregression

The current feature CLI writes *_descriptors.csv, *_mordred.csv, and *_morgan.csv. Some older checked-in processed files in this repo still use the legacy *_rdkit.csv naming.

Documentation

• GitHub Pages: https://forecasth.github.io/ligand-classification/
• Docs source index: docs/README.md
• Key guides:
  • Getting started
  • Data preparation
  • Training pipeline
  • CLI reference
  • Metrics
  • Development

Repository outline

• src/cli/: feature generation, exploration, training, evaluation, and prediction commands
• src/training/: dataset preparation, pipeline assembly, model registry, and reporting
• src/features/: RDKit and DeepPK helpers
• configs/: starter configs and model hyperparameter grids
• docs/: MkDocs source for the GitHub Pages site

Verification

The repo test suite currently passes with pytest -q. The docs site is built with mkdocs build --strict.