gcorso/DiffDock

Leverages equivariant diffusion models to jointly predict ligand position, orientation, and conformational changes in 3D space, with built-in confidence scoring for pose quality assessment. Supports flexible input formats including PDB files, protein sequences (auto-folded with ESMFold), SMILES strings, and batch processing via CSV, running efficiently on GPU or CPU. DiffDock-L, the latest version, significantly improves generalization and performance; the framework integrates with standard cheminformatics tools (RDKit) and provides deployment options including Hugging Face Spaces, local UI, and Docker containers.

1,454 stars. No commits in the last 6 months.