GitHub - lanl/minervachem: a python library for cheminformatics and machine learning

Abstraction: LANL Python library for graphlet molecular fingerprints and interpretable ML

Key points:

• Implements Graphlet Fingerprints: fragments identified by atom count and hash; reusable GraphletFingerprinter objects
• Provides sklearn-compatible FingerprintFeaturizer for building sparse feature matrices from molecule sets
• Supports hierarchical residual linear models for interpretable property prediction (tested on ~100k QM9 molecules with ~6,000+ coefficients)
• DAG-based projection enables atom- and bond-level visualization of model coefficient contributions
• Alpha release (breaking changes expected); conda install recommended over pip; paper on ChemRxiv (doi:10.26434/chemrxiv-2024-r81c8)
• BSD-3 licensed; developed at Los Alamos National Laboratory under DOE contract

Connections: Los Alamos National Laboratory · Cheminformatics · Molecular Fingerprints · Interpretable Machine Learning

Source: https://github.com/lanl/minervachem