Deep reinforcement learning for protein complex modeling

Dependencies

• PyRosetta==4
• numpy>=1.20.2
• pandas>=1.2.5
• tqdm==4.61.1
• scipy>=1.6.2
• seaborn>=0.11.2
• setuptools>=44.0.0
• imageio>=2.10.1
• matplotlib>=3.4.2
• tensorflow==1.15

Installation

The package is tested using Python 3.6 and 3.7. To install the software, please follow the instructions below:

• Install the above dependencies
• Download and install PyRosetta (http://www.pyrosetta.org/dow)
• Install the package following the instruction below:

git clone git@github.com:jianlin-cheng/DeepRLP.git

(If fail, try username) git clone https://github.com/jianlin-cheng/DeepRLP.git

cd DeepRLP
pip install -r requirements.txt


Alternatively, environment.yml files are provided to install the required packages using pip or conda.

Basic Usage

1. Reconstructing the dimer structure using true structure

python ./scripts/true_structure/dqn_docking.py <initial_pdb> <native_strcuture>

2. Reconstructing the dimer structure using true/predicted contacts

python ./scripts/predicted_contacts/dqn_docking.py <path_to_the_DeepRLP_tool> <initial_start> <res_file> <out_dir> <target_name>

 

A video demonstrating how DRLComplex reconstructs the dimer structure using true interchain contacts:

drl_final_rewards_faster.mp4