PyTorch implementation of MolDQN as described in Optimization of Molecules via Deep Reinforcement Learning by Zhenpeng Zhou, Steven Kearnes, Li Li, Richard N. Zare and Patrick Riley.

1. Install rdkit.
   conda create -c rdkit -n my-rdkit-env rdkit
conda activate my-rdkit-env
conda install -c conda-forge rdkit

2. Clone this repository.
   git clone https://github.com/aksub99/MolDQN-pytorch.git
cd MolDQN-pytorch

3. Install the requirements given in requirements.txt.
   pip install -r requirements.txt

4. Install baselines.
   pip install "git+https://github.com/openai/baselines.git@master#egg=baselines-0.1.6"

Using a docker image requires an NVIDIA GPU. If you do not have a GPU please follow the directions for installing from source In order to get GPU support you will have to use the nvidia-docker plugin.

# Build the Dockerfile in Dockerfiles/Dockerfile to create a Docker image.
cd Dockerfiles
docker build -t moldqn_pytorch:latest .

# This will create a container from the image we just created.
nvidia-docker run -[Options] moldqn_pytorch:latest python path/to/main.py

Please remember to modify the TB_LOG_PATH variable in main.py depending on where you wish to store your tensorboard runs file.

python main.py

A simple example to train the model on a single property optimization task can be seen in examples/MolDQN-pytorch.ipynb.

The following was the reward curve obtained when the model was trained for 5000 episodes on a single property optimization task (QED in this case).