DiffDock — молекулярный докинг с диффузионными моделями

★ 7.0 · data

diffdock is a Claude Code skill that runs molecular docking using the DiffDock and DiffDock-L diffusion-based deep learning models, predicting 3D binding poses of small molecule ligands against protein targets. It accepts protein structures as PDB files or amino acid sequences processed via ESMFold, alongside ligands in SMILES, SDF, or MOL2 formats, and supports both single-complex docking and large-scale virtual screening with pre-computed ESM embeddings. Each run produces ranked pose files with per-pose confidence scores embedded in filenames; crucially, DiffDock predicts binding geometry rather than binding affinity (ΔG, Kd), which requires separate scoring tools such as GNINA or MM/GBSA. The environment targets DiffDock v1.1.3, requiring Python 3.9, RDKit, PyTorch/PyG, and the DiffDock repository, with optional CUDA GPU acceleration for 10–100× speedup. The skill suits computational chemists and structural biologists who need to automate protein-ligand docking pipelines directly within Claude Code.