molfeat

You convert SMILES strings into numerical features for machine learning by using configurable molecular featurizers that generate fingerprints and descriptors into a NumPy matrix. This provides consistent, ML-ready input for downstream QSAR modeling.

Yes, you can integrate molecular featurization directly into scikit-learn pipelines for virtual screening. The featurization process uses scikit-learn compatible transformers that support parallel processing, caching, and error-tolerant batch operations.

For QSAR modeling, you can generate unified molecular features including ECFP fingerprints, MACCS keys, 2D/3D descriptors, pharmacophore features, and pretrained embeddings from a single interface.

Molecular featurization supports reproducible configurations for model deployment by allowing you to save and reload the exact featurizer and transformer state, ensuring consistent training and deployment across batches.

You process a large compound library for similarity searching without errors by using error-tolerant batch processing with parallelization. This allows the featurization transformers to handle large datasets efficiently while skipping problematic entries.

You do not strictly need RDKit molecules to compute 2D and 3D descriptors, as the featurization accepts standard SMILES strings directly. However, it also fully supports RDKit molecules as input for generating numerical features.