externelly/diffdock repository overview

⁠Start the DiffDock Docker image

This Docker image includes the per-trained DiffDock and ESM models and it takes 18.54gb of storage after downloading. Follow the instructions below to run it using either CPU or CUDA-enabled GPU. The Docker image is of DiffDock-L (v1.1) that was built on 04.15.24 with the latest updates through GitHub⁠.

docker pull externelly/diffdock:latest

# To run DiffDock with CPU only
docker run --name=diffdock -v ${PWD}/mount:/DiffDock/mount --rm -it externelly/diffdock:latest

# To run DiffDock with CUDA GPU enabled
docker run --gpus all --name=diffdock -v ${PWD}/mount:/DiffDock/mount --rm -it externelly/diffdock:latest

⁠Preparation of the input file

Create a pdbs/ folder in your mount folder and copy all PDB files into it.

Prepare a mount/diffdock_analysis.csv CSV file in the mount folder with the following columns:

protein_path,ligand_description,complex_name,protein_sequence
mount/pdbs/PDB-id.pdb,SMLES,PDB-id_SMILES-id,
...
# one row for each PDB-SMILES combination

⁠Inside the docker image

Run the followin command to download models: python -m inference.
It will takes a couple of minutes. Not persistant once you close the docker container.

⁠Start the analysis

python -m inference --config default_inference_args.yaml --protein_ligand_csv mount/diffdock_analysis.csv --out_dir mount/results

⁠Cite

Don't forget to cite the original paper.