Installation

Option 1 - one-line install from scratch with conda:

conda env create -f environment.yml && conda activate qsarmil

Option 2 - install into an existing environment:

pip install qsarmil

Beginner usage

For a predefined pipeline that takes zero QSAR/MIL knowledge and zero setup: hand it your training data and get predictions for new molecules back.

from qsarmil.modelling.meta import MultiConformerRegressor

smiles_train = [
    "CC(C)Cc1ccc(cc1)C(C)C(=O)O",
    "COc1ccc2cc(ccc2c1)C(C)C(=O)O",
    "OC(=O)Cc1ccccc1Nc1c(Cl)cccc1Cl",
    "COc1ccc2c(c1)c(CC(=O)O)c(C)n2C(=O)c1ccc(Cl)cc1",
    "OC(=O)C(C)c1cccc(c1)C(=O)c1ccccc1",
]
y_train = [5.2, 5.9, 6.1, 7.0, 5.6]

smiles_test = ["CC(C(=O)O)Oc1cccc(c1)-c1ccccc1", "CC(C(=O)O)c1ccc(cc1)-c1ccc(F)cc1"]

# train and predict in one call - there's no separate save/load step
model = MultiConformerRegressor(num_conf=10, hopt=True, verbose=True)
y_pred = model.train_predict(smiles_train, y_train, smiles_test)

Use MultiConformerRegressor for continuous properties and MultiConformerClassifier for binary classification. See the full walkthrough in 01_Beginner_Easy_Start.ipynb.

Professional usage

Modify or build your own modeling pipeline by combining QSARmil’s individual modules (conformer generator, descriptor calculator, multi-instance methods, etc.) directly. See 02_Professional_Pipeline_Customization.ipynb for the details.

Tutorials

• 01_Beginner_Easy_Start.ipynb - the predefined, zero-configuration pipeline.

• 02_Professional_Pipeline_Customization.ipynb - building a custom pipeline from individual modules.

• 03_Key_Instance_Detection.ipynb - identifying which conformers drive a prediction.

Documentation

Full documentation is available at QSARmil docs.