Source code for the Nature Communications paper DynamicBind: predicting ligand-specific protein-ligand complex structure with a deep equivariant generative model.

DynamicBind recovers ligand-specific conformations from unbound protein structures (e.g. AF2-predicted structures), promoting efficient transitions between different equilibrium states.

Create environment dynamicbind:

conda create -n dynamicbind python=3.10
conda activate dynamicbind
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.4.0+cu121.html
pip install torch_geometric -f https://data.pyg.org/whl/torch-2.4.0+cu121.html
pip install rdkit
pip install pyyaml biopython
pip install e3nn fair-esm spyrmsd
pip install pandas
pip install tqdm

Create another environment for structural Relaxation.

conda create --name relax python=3.8
conda activate relax
pip install openmm
conda install -c conda-forge pdbfixer
pip install compilers biopython
pip install tqdm
pip install scipy
pip install pandas
pip install rdkit
pip install networkx

mkdir esm_models
cd esm_models/
mkdir checkpoints
cd checkpoints/
wget -c https://dl.fbaipublicfiles.com/fair-esm/models/esm2_t33_650M_UR50D.pt
wget -c https://dl.fbaipublicfiles.com/fair-esm/regression/esm2_t33_650M_UR50D-contact-regression.pt

Download and unzip the workdir.zip containing the model checkpoint form https://zenodo.org/records/10137507, v2 is contained here https://zenodo.org/records/10183369.

By default: 40 poses will be predicted, poses will be ranked (rank1 is the best-scoring pose, rank40 the lowest), relax processes are included.

1. Protein (PDB File): protein.pdb
   • Automatically cleaned to remove non-standard amino acids, water molecules, or small molecules.
2. Ligand (CSV File): ligand.csv
   • Must contain a column named 'ligand' listing smiles.
3. Number of Animations:
   • outputs intermediate pkl data, not the final animation PDB. (After --savings_per_complex, default is 40)
4. Frames in Animation/inference_steps:
   • default is 20.

• --header: Name of the result folder.
• --device: GPU device ID.
• --python: Python environment for inference.
• --relax_python: Python environment for relaxation.
• --num_workers: Number of processes for final output relaxation.

python run_single_protein_inference.py data/origin-1qg8.pdb data/1qg8_input.csv \
--savings_per_complex 40 \
--inference_steps 20 \
--header test --device 0 \
--python /home/ruofan/anaconda3/envs/dynamicbind/bin/python \
--relax_python /home/ruofan/anaconda3/envs/relax/bin/python \
--movie