Skip to content

Latest commit

 

History

History

AbDock

README.md

Pratope-epitope Docking using AbDock

Table of Contents

About

This is the code implementation of AbDock, a generative diffusion model-based paratope-epitope docking method. We also integrate the code of the antibody optimization pipeline for reproducing the optimization results in the paper.

Getting Started

Installing

The code relies on PyTorch with cuda support. We recommend using conda to install the dependencies.

conda create -n antibody_dock python=3.8 -y
conda activate antibody_dock
conda install pytorch==1.11.0 cudatoolkit=11.3 -c pytorch -yconda install pyg::pytorch-scatter=2.0.9 -y
conda install pyyaml mmseqs2  -y 
conda install -c conda-forge pdbfixer -y
conda install -c bioconda abnumber -y
pip install -U "ray[default]"
pip install dynamic-yaml tqdm biopython pandas joblib lmdb easydict scipy wandb notebook seaborn
Install pyrosetta.

Please obtain the username and password on the PyRosetta website.

wget --user={user} --password={psw} https://graylab.jhu.edu/download/PyRosetta4/archive/release/PyRosetta4.Release.python38.linux/PyRosetta4.Release.python38.linux.release-350.tar.bz2
tar -xvjf PyRosetta4.Release.python38.linux.release-350.tar.bz2
cd PyRosetta4.Release.python38.linux.release-350
python setup/setup.py install
cd ..

Usage

Paratope-epitope Docking using AbDock

To perform docking for pdb_file data/examples/7DK2_AB_C.pdb and save the results to ./results/dock_cdr/7DK2_AB_C.pdb_.

python dock_pdb.py --pdb_path data/examples/7DK2_AB_C.pdb
  • 7DK2_AB_C.pdb has the A (heavy chain) and B (light chain) chains of the antibody and the C chain (epitope) of the antigen.
  • The script will automatically identify the CDRH3 and perform docking to the antigen.
  • Please refer to src/tools/runner/design_for_pdb.py:args_from_cmdline for complete arguments.
  • Please check out ./results/dock_cdr/7DK2_AB_C.pdb_ for the docking results.

Antibody Optimization Pipeline

The optimization pipeline consists of three steps:

  1. Pose generation using AbDock.
  2. Desgin sequences for generated poses.
  3. Sequence screening using AbDock.

Generate four point-mutation results for 7bsd_A_B_G

Run AbDock to generate docking poses from native CDRH3 sequence.

python dock_pdb.py --pdb_path data/examples/7bsd_A_B_G.pdb --config configs/test/dock_cdr.yml --ckpt reproduction/dock_single_cdr/250000.pt -n 1000 -b 1000 -d "cuda" -o results

Run the sequence design and screening pipeline. The four points are specified by --design_contig "6-9".

python optimize_ab.py --num_gpus 1 --process_per_gpu 1 --docked_pose_dir "results/dock_cdr/7bsd_A_B_G.pdb_/H_CDR3" --seq_design_dir "results/seq_design_fixed_pos/mutation/CDRH3_7_9" --design_model_ckpt "reproduction/seq_design_fixed_pos/300000.pt" --design_contig "6-9" --screen_dir "results/screening/seq_design_fixed_pos/mutation/CDRH3_7_9" --dock_model_ckpt "reproduction/dock_single_cdr/250000.pt" --heavy_chain_id "A" --nums 100 --pdb_suffix "rosetta"

Run the notebook ab_opt_analysis_4mutations.ipynb to analyse the optimiaztion results.

Reproduce Training

Download and preprocess the dataset

Download Protein structures in the SAbDab. Extract all_structures.zip into the data folder.

Launch training

python train.py configs/train/dock_single.yml
  • Please refer to train.py for complete arguments.
  • Please check out configs/train for the list of training setting.