Docking: DiffDock-L
A diffusion-based molecular docking lab built on DiffDock-L v1.1.3. Provide a receptor PDB and a ligand SMILES (or RDKit-readable file), get back diffusion-sampled poses ranked by confidence.
What it simulates
- DiffDock-L inference in managed runtime mode: clones the pinned upstream repo, builds a venv, caches both for later runs.
- Multiple diffusion samples scored with a confidence model.
- Bundled
1a0qreceptor and SMILES ligand so a fresh run docks without setup. - Pose table, confidence summary, and the merged top-ranked complex.
Run it on the Hub
- Open the DiffDock: DiffDockLDockingPredictor Lab on the public Hub.
- Click Run. The bundled
1a0qdefaults dock without any parameter editing.
Inputs you can tune
| Input | Meaning |
|---|---|
protein_path | Receptor PDB file. Defaults to data/1a0q/1a0q_protein_processed.pdb. |
ligand_description | SMILES string or path to an RDKit-readable ligand file. |
run_options.samples_per_complex | Number of DiffDock samples to draw. |
run_options.inference_steps | Denoising steps per sample. |
run_options.batch_size | Inference batch size. |
run_options.save_visualisation | Whether to write per-step reverse-process PDBs. |
What results to expect
- Pose ranking table: rank, DiffDock confidence (unitless log-likelihood-style score), confidence band (
high,moderate, orlow), and the SDF filename. - 3D structure view: receptor with the top pose merged in as
top_rank_complex.pdb. - Confidence summary: top-pose score, band, total pose count, and every per-pose confidence so bimodal pose distributions are visible.
- Run metadata: runtime mode, cache directories, executed command, returncode, truncated stdout/stderr, and
status: completed/status: error.
The lab exposes generated PDB and SDF files through structure_artifacts, so remote runs upload them as durable run artifacts before completion.
Source on GitHub: models-diffdock. See Artifact Outputs for the file-output contract. The first run needs internet access (clones the upstream repo and downloads model checkpoints); later runs are offline.