Getting Started¶
Prerequisites¶
- Conda is the easiest way to install RDKit and the rest of the scientific stack.
- Python 3.11 is the pinned runtime in
environment.yml. - Internet access is only required for DeepPK submissions and package installation.
Create the environment¶
The environment file includes:
- chemistry and modeling packages such as
rdkit,scikit-learn,xgboost,lightgbm,imbalanced-learn,mordred, andumap-learn - tooling such as
pytest,requests,tqdm,mkdocs, andmkdocs-material
First commands¶
Generate current-format example features from the checked-in raw example:
python -m src.cli.process_features data/example/raw/psychlight_a.csv \
--output-dir data/example/processed/psychlight_a
Run a narrow training pass with the canonical default config:
python -m src.cli.train --config configs/train.yaml \
--datasets psychlight_a_descriptors \
--models logisticregression
The starter config expects:
data/example/processed/psychlight_a/psychlight_a_descriptors.csvdata/example/processed/psychlight_a/psychlight_a_mordred.csvdata/example/processed/psychlight_a/psychlight_a_morgan.csv
Optional DeepPK flow¶
DeepPK is a third-party web service. When you enable it, the processed SMILES are sent to a remote server rather than computed locally.
python -m src.cli.process_features data/example/raw/psychlight_a.csv \
--output-dir data/example/processed/psychlight_a \
--deeppk-post \
--deeppk-pred-type admet \
--deeppk-timeout 3600
Then train with the DPK-aware config:
Verify the installation¶
Run the repository checks that matter for this project:
pytest -q
conda run -n lig-cls python -m src.cli.process_features --help
conda run -n lig-cls python -m src.cli.train --help
conda run -n lig-cls python -m src.cli.evaluate --help
conda run -n lig-cls python -m src.cli.predict --help
conda run -n lig-cls python -m src.cli.explore --help
mkdocs build --strict
If python -m src.cli.process_features --help fails outside the project environment, that usually means RDKit is missing from the active interpreter.