Deepchem

DeepChem brings serious molecular ML infrastructure to Claude. You get property prediction for ADMET and toxicity, graph neural networks like GCN and MPNN, and direct access to MoleculeNet benchmarks with proper scaffold splitting to prevent data leakage. The featurization layer alone saves weeks of work, converting SMILES strings and SDFs into fingerprints or graph representations depending on your model choice. Pretrained models like ChemBERTa and GROVER are available for transfer learning when your dataset is small. If you're doing drug discovery ML or materials informatics and tired of stitching together RDKit, PyTorch Geometric, and custom loaders, this consolidates the stack. The API is opinionated about best practices, which is exactly what you want in chemistry ML where it's easy to leak signal.