DiffDock-L

Diffusion-based protein-ligand docking with SMILES, SDF, or CSV batches.

Basics

Entry point: python -m inference; supply one flag per line.
Provide either --protein_ligand_csv or both --protein_path and --ligand_description.
Common flags: --protein_path, --ligand_description (SMILES or ligand file path), --protein_ligand_csv (batch), --out_dir, --samples_per_complex, --inference_steps, --save_visualisation.
The service forces --model_dir=/ref/workdir/v1.1/score_model (user overrides ignored).
If --out_dir is omitted, the service uses /outputs/diffdock_l.

Straightforward run

--protein_path=/inputs/1a46_protein_processed.pdb
--ligand_description=/inputs/1a46_ligand.sdf
--complex_name=run1
--out_dir=/outputs/diffdock_l_run1
--samples_per_complex=1
--inference_steps=5
--save_visualisation

Expected outputs: ranked ligand rank*.sdf. With --save_visualisation, you also get rank*_reverseprocess.pdb.

To view a full complex, load receptor PDB + ranked ligand SDF together in PyMOL/Chimera.

Submit

Launch via New Job -> DiffDock-L; the form pre-fills the SMILES example.