Integrated structure prediction of protein–protein docking with experimental restraints using ColabDock
ColabDock is a docking framework developed based on ColabDesign. It is able to incorporate experimental restraints to generate accurate protein complex structure. For details, please refer to the paper.
News
- July 5, 2024
- We are excited to announce that ColabDock has been published on Nature Machine Intelligence 😆! (Link)
Note running ColabDock locally requires GPU. If you do not have one, we suggest using the Colab version.
Table of content
- Create a python enviroment using conda.
conda create --name colabdock python=3.8- Clone the repo
git clone git@github.com:JeffSHF/ColabDock.git- Activate environment
cd ColabDock && conda activate colabdock- Install jax
Please refer to JAX github page to install package corresponding to your CUDA and cudnn version.
Example (which has been tested locally):
# install jaxlib
pip install https://storage.googleapis.com/jax-releases/cuda11/jaxlib-0.3.8+cuda11.cudnn805-cp38-none-manylinux2014_x86_64.whl
# install jax
pip install jax==0.3.8- Install other dependencies
pip install -r requirements.txt- Download AlphaFold2 parameters
mkdir params
cd params
wget https://storage.googleapis.com/alphafold/alphafold_params_2022-12-06.tar
tar -xvf alphafold_params_2022-12-06.tarThe installation takes ~10 min.
Before running the code, please set the variables in the config.py file according to the descriptive information in it. The default config.py contains the setting of protein 4INS4.
conda activate colabdock && python main.pyThe running time depends on the size of the protein, the round and step numbers. For the default setting, it should take ~10 min.
After the running, the outputs directory (default is results) will contain three folders, i.e., gen, pred, and docked. Gen and pred contain the pdb files of all the docking structures derived from the generation stage and the prediction stage. The docked folder contains the top5 predicted docking structures.
If you want to test ColabDock using a complex with known structure, you can generate 1v1, 1vN, or MvN restraints using the extract_rest.py script. For example, 4INS4 contains 4 protein chains, i.e., A,B,C,D. If you want to sample some 1v1 restraints between chain A and chain D, then run the following command and the program will print the sampled restraints.
python extract_rest.py ./protein/4INS4/PDB/native.pdb A,B,C,D A,D 1v1You can also use python extract_rest.py -h to get more informations about the input parameters.
Suppose the complex you want to dock contains two chains, i.e., A and B, and each chain contains 10 amino acids. The "chains" variable in the config.py file is set to 'A,B'.
If you want the 4th residue in chain A close to the 5th residue in chain B in the docking structure, the 1v1 restraint (rest_1v1 variable in the config.py file) should be set to [4, 15].
If you want the 4th residue in chain A close to one of residues from 3rd to 7th in chain B in the docking structure, the 1vN restraint (rest_1vN variable in the config.py file) should be set to [4, range(13, 18)].
If you have two 1vN restraints, i.e., [4, range(13, 18)] and [6, range(13, 18)], and you want at least one of them satisfied in the docking structure, the MvN restraint (rest_MvN variable in the config.py file) should be set to [[4, range(13, 18)], [6, range(13, 18)], 1].
If you want the 6th residue in chain A far away from the 8th residue in chain B in the docking structure, the repulsive restraint (rest_rep variable in the config.py file) should be set to [6, 18].
In ColabDock paper, we sampled restraints for proteins collected from protein docking benchmark 5.5. The sampled restraints can be downloaded from OSF.
A blog in Chinese (link)
- Shihao Feng @JeffSHF
- Zhenyu Chen @Dreams1de
- Sergey Ovchinnikov @sokrypton