alterlab-molfeat

Molecular featurization converts SMILES or RDKit molecules into ML-ready vectors using fingerprints, descriptors, or pretrained models. You can apply batch processing via MoleculeTransformer to standardize feature spaces for QSAR modeling pipelines.

Yes, you can concatenate features using FeatConcat to combine pretrained transformer embeddings and RDKit descriptors. This unified feature space is suitable for large-scale virtual screening and deep-learning pipelines.

Batched molecular featurization is handled by MoleculeTransformer, which processes lists of molecular structures into consistent vector representations. It seamlessly integrates with scikit-learn pipelines for deep-learning workflows.

Yes, the framework supports a wide range of featurizers including MAP4 fingerprints and Mordred descriptors. You can use the ModelStore to discover, load, and compare these specific featurizers for your clustering tasks.

Yes, advanced featurizers require optional dependencies to function properly. The core framework provides standard fingerprints and descriptors, while advanced pretrained models and specific embeddings need additional setup.

A unified framework streamlines data-driven discovery by providing a single ModelStore for discovering and reusing featurizers. It prevents alignment issues and enables consistent feature concatenation across different molecular representations.