Torchdrug

If you're doing ML on molecules or proteins, this is a solid PyTorch toolkit that handles the graph structure natively. It includes 40+ datasets (BBBP, Tox21, QM9 for molecules; EnzymeCommission, PDBBind for proteins), 20+ GNN architectures (GIN, GAT, SchNet), and pre-built tasks for property prediction, retrosynthesis, and molecular generation. The knowledge graph reasoning over biomedical data like Hetionet is surprisingly useful for drug repurposing. It integrates well with RDKit and plays nice with pre-trained protein models like ESM. The scaffold splitting and property prediction workflows are production-ready, though you'll want the full installation for all features.