scikit-fingerprints

scikit-fingerprints is a Python library for efficient computation of molecular fingerprints.

Table of Contents

• Description
• Supported platforms
• Installation
• Quickstart
• Project overview
• Contributing
• License

Description

Molecular fingerprints are crucial in various scientific fields, including drug discovery, materials science, and chemical analysis. However, existing Python libraries for computing molecular fingerprints often lack performance, user-friendliness, and support for modern programming standards. This project aims to address these shortcomings by creating an efficient and accessible Python library for molecular fingerprint computation.

You can find the documentation HERE

Main features:

• scikit-learn compatible
• feature-rich, with >30 fingerprints
• parallelization
• sparse matrix support
• commercial-friendly MIT license

Supported platforms

Installation

You can install the library using pip:

pip install scikit-fingerprints

If you need bleeding-edge features and don't mind potentially unstable or undocumented functionalities, you can also install directly from GitHub:

pip install git+https://github.com/scikit-fingerprints/scikit-fingerprints.git

Quickstart

Most fingerprints are based on molecular graphs (2D-based), and you can use SMILES input directly:

from skfp.fingerprints import AtomPairFingerprint

smiles_list = ["O=S(=O)(O)CCS(=O)(=O)O", "O=C(O)c1ccccc1O"]

atom_pair_fingerprint = AtomPairFingerprint()

X = atom_pair_fingerprint.transform(smiles_list)
print(X)

For fingerprints using conformers (3D-based), you need to create molecules first and compute conformers. Those fingerprints have requires_conformers attribute set to True.

from skfp.preprocessing import ConformerGenerator, MolFromSmilesTransformer
from skfp.fingerprints import WHIMFingerprint

smiles_list = ["O=S(=O)(O)CCS(=O)(=O)O", "O=C(O)c1ccccc1O"]

mol_from_smiles = MolFromSmilesTransformer()
conf_gen = ConformerGenerator()
fp = WHIMFingerprint()
print(fp.requires_conformers)  # True

mols_list = mol_from_smiles.transform(smiles_list)
mols_list = conf_gen.transform(mols_list)

X = fp.transform(mols_list)
print(X)

You can also use scikit-learn functionalities like pipelines, feature unions etc. to build complex workflows. Popular datasets, e.g. from MoleculeNet benchmark, can be loaded directly.

from skfp.datasets.moleculenet import load_clintox
from skfp.metrics import multioutput_auroc_score
from skfp.model_selection import scaffold_train_test_split
from skfp.fingerprints import ECFPFingerprint, MACCSFingerprint
from skfp.preprocessing import MolFromSmilesTransformer

from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import make_pipeline, make_union


smiles, y = load_clintox()
smiles_train, smiles_test, y_train, y_test = scaffold_train_test_split(
    smiles, y, test_size=0.2
)

pipeline = make_pipeline(
    MolFromSmilesTransformer(),
    make_union(ECFPFingerprint(count=True), MACCSFingerprint()),
    RandomForestClassifier(random_state=0),
)
pipeline.fit(smiles_train, y_train)

y_pred_proba = pipeline.predict_proba(smiles_test)
auroc = multioutput_auroc_score(y_test, y_pred_proba)
print(f"AUROC: {auroc:.2%}")

Project overview

scikit-fingerprint brings molecular fingerprints and related functionalities into the scikit-learn ecosystem. With familiar class-based design and .transform() method, fingerprints can be computed from SMILES strings or RDKit Mol objects. Resulting NumPy arrays or SciPy sparse arrays can be directly used in ML pipelines.

Main features:

1. Scikit-learn compatible: scikit-fingerprints uses familiar scikit-learn interface and conforms to its API requirements. You can include molecular fingerprints in pipelines, concatenate them with feature unions, and process with ML algorithms.

2. Performance optimization: both speed and memory usage are optimized, by utilizing parallelism (with Joblib) and sparse CSR matrices (with SciPy). Heavy computation is typically relegated to C++ code of RDKit.

3. Feature-rich: in addition to computing fingerprints, you can load popular benchmark datasets (e.g. from MoleculeNet), perform splitting (e.g. scaffold split), generate conformers, and optimize hyperparameters with optimized cross-validation.

4. Well-documented: each public function and class has extensive documentation, including relevant implementation details, caveats, and literature references.

5. Extensibility: any functionality can be easily modified or extended by inheriting from existing classes.

6. High code quality: pre-commit hooks scan each commit for code quality (e.g. black, flake8), typing (mypy), and security (e.g. bandit, safety). CI/CD process with GitHub Actions also includes over 250 unit and integration tests.

Contributing

Please read CONTRIBUTING.md and CODE_OF_CONDUCT.md for details on our code of conduct, and the process for submitting pull requests to us.

Citing

If you use scikit-fingerprints in your work, please cite our main publication, available on SoftwareX (open access):

@article{scikit_fingerprints,
   title = {Scikit-fingerprints: Easy and efficient computation of molecular fingerprints in Python},
   author = {Jakub Adamczyk and Piotr Ludynia},
   journal = {SoftwareX},
   volume = {28},
   pages = {101944},
   year = {2024},
   issn = {2352-7110},
   doi = {https://doi.org/10.1016/j.softx.2024.101944},
   url = {https://www.sciencedirect.com/science/article/pii/S2352711024003145},
   keywords = {Molecular fingerprints, Chemoinformatics, Molecular property prediction, Python, Machine learning, Scikit-learn},
}

Its preprint is also available on ArXiv.

License

This project is licensed under the MIT License - see the LICENSE.md file for details.