A web-based platform that predicts druggable binding pockets with P2Rank, docks single ligands or whole libraries with AutoDock Vina, generates flexible-receptor ensembles via NMA or short OpenMM MD, computes ADMET properties with RDKit, optionally refines poses with OpenMM, and rescores with an approximate MM-GBSA method — all visualized in your browser with 3Dmol.js.
Documentation: https://pocketdock.readthedocs.io/
- Pocket detection — automatic druggable-pocket prediction with P2Rank v2.5
- Single and batch docking — submit one ligand or up to 100 in a single batch
- Ensemble docking — generate 2–10 receptor conformations via NMA (fast) or OpenMM MD (thorough)
- ADMET properties — RDKit-computed MW, logP, HBA/HBD, TPSA, QED, Lipinski & Veber pass/fail (every job)
- Pose refinement (optional) — OpenMM AMBER14 + OBC2 implicit-solvent energy minimization
- MM-GBSA-style rescoring (optional) — per-pose ΔG (kJ/mol), sortable in the results table
- Interaction analysis — H-bonds, hydrophobic, salt bridges, π-stacking, π-cation, halogen bonds
- Interactive 3D viewer — 3Dmol.js with sortable results, 2D interaction maps, CSV / PNG export
- REST API + Celery queue — script everything from Python or curl
Option A — run the published image (no clone needed): copy the compose snippet from Run from the published image into a new directory and run docker compose up.
Option B — build from source:
git clone https://github.com/gozsari/PocketDock.git
cd PocketDock
docker compose up --buildThe application will be available at http://localhost:8000 in either case. See the Getting Started guide for prerequisites and first-run walkthrough.
A ready-to-run example is shipped in examples/egfr_erlotinib/ — the EGFR kinase domain (PDB 4HJO) plus erlotinib. Upload the two files on the home page, submit, and you should recover the canonical ATP-pocket binding mode in a few minutes.
| Service | Description |
|---|---|
| web | Django app serving the UI and REST API |
| celery | Celery worker running the docking pipeline |
| redis | Message broker and result backend |
- (Optional) Ensemble — generate N receptor conformations with NMA or short OpenMM MD
- P2Rank detects binding pockets and ranks them by druggability
- Meeko prepares receptor and ligand PDBQT files
- RDKit computes ADMET descriptors from the ligand
- AutoDock Vina docks the ligand into the top N pockets
- Interaction analysis annotates each pose with detected contacts
- (Optional) OpenMM refinement energy-minimizes each pose
- (Optional) MM-GBSA rescoring computes a per-pose ΔG
Results are ranked by a combined score of pocket probability and binding affinity. See Interpreting Results for the full scoring details.
Environment variables (set in docker-compose.yml):
| Variable | Default | Description |
|---|---|---|
VINA_EXHAUSTIVENESS |
8 |
Vina search exhaustiveness |
VINA_NUM_MODES |
9 |
Max number of docking poses per pocket |
VINA_BOX_PADDING |
5.0 |
Padding around the pocket (Å) |
VINA_DEFAULT_BOX_SIZE |
20.0 |
Fallback grid-box edge length (Å) |
WORKER_CONCURRENCY |
2 |
Celery worker processes |
Full reference: Configuration.
- Backend — Django 5.x, Django REST Framework, Celery 5.4+
- Docking — P2Rank 2.5, AutoDock Vina 1.2.7, Meeko 0.5+
- Cheminformatics — RDKit (ADMET, ligand parsing)
- Structural mechanics — OpenMM + PDBFixer (ensemble MD, pose refinement); Gemmi (structure I/O)
- Frontend — Django templates, Tailwind CSS, 3Dmol.js
- Infrastructure — Docker Compose, Redis 5.0+, Gunicorn
| Method | Endpoint | Description |
|---|---|---|
| GET | / |
Upload page (single + batch tabs) |
| POST | /api/jobs/ |
Create a docking job |
| GET | /api/jobs/<id>/status/ |
Poll job status |
| GET | /api/jobs/<id>/results/ |
Get results (poses, ADMET, MM-GBSA) |
| GET | /api/jobs/<id>/files/<path> |
Serve molecular files |
| GET | /api/batch/<batch_id>/ |
Batch progress + per-ligand best scores |
| GET | /api/ensemble/<ensemble_id>/ |
Ensemble progress + consensus top-20 |
| GET | /jobs/<id>/ |
Job status / results page |
| GET | /batch/<batch_id>/ |
Batch dashboard page |
| GET | /ensemble/<ensemble_id>/ |
Ensemble dashboard page |
Full reference with request/response schemas: API Reference.
If you use PocketDock in your research, please cite it using the metadata in CITATION.cff (v1.0.0, released 2026-05-20).
The author thanks the developers of P2Rank, AutoDock Vina, OpenMM, PDBFixer, RDKit, Meeko, 3Dmol.js, NumPy and SciPy, whose open-source tools PocketDock builds upon. Development of the codebase was assisted in part by the Claude large language model (Anthropic) as a coding aid.