Pocket2Mol_RL_public

Public repository of the paper <Fine-tuning Pocket-conditioned 3D Molecule Generation via Reinforcment Learning>

License

This work © 2024 by Deargen Inc. is licensed under CC BY-NC-SA 4.0. Please keep the LICENSE file and the current section of README.md as they are.

Note that a provisional patent is under review, which potentially limits the patentability of related ideas.

How to reproduce the results of the paper

Reinforcement Learning

We do not provide the code for training models.

Environment Setup

1. This guide outlines the steps to set up your environment using the Docker image deargen/pocket2mol_rl_public:latest. You can work with this environment either interactively in a terminal or within a development container in Visual Studio Code.

2. If you don't want to use Docker, you can set up the environment using the Conda environment file environment.yml and install the package from the current repository.

3. Choose one of the following options to set up your environment and proceed to the finally section to install the current repository.

Option1) Using Docker Interactively

Open your terminal and execute the command below to start an interactive bash shell within the Docker container:

docker run -it deargen/pocket2mol_rl_public:latest /bin/bash

This will pull the deargen/pocket2mol_rl_public:latest image from DockerHub (if not already present locally) and open a bash shell in the container.

You have to logged in to the DockerHub to pull the image.

Option2) Using Visual Studio Code Dev Containers

To open a folder in a dev container using Visual Studio Code, ensure you have the Remote - Containers extension installed, then follow these steps:

• Press Command + Shift + P on macOS (Ctrl + Shift + P on Windows/Linux) to open the command palette.
• Type and select Dev Container: Open Folder in Container.
• Follow the prompts to select the folder you wish to open in the container.
• For more details, refer to the Visual Studio Code documentation on dev containers.

Option3) Setting Up Environment from Conda Environment File

1. Create the Conda Environment: Execute the following command to create the Conda environment:

   conda env create -f environment.yml

2. Activate the Conda Environment: Once the environment is successfully created, activate it using:

   conda activate pocket2mol_rl

   Replace pocket2mol_rl with the name of your environment.

3. Install Openbabel manually This command installs the openbabel package for the user, independent of any Conda environment. Openbabel is a dependency to compute docking scores. Because the openbabel package is not available in the conda environment, you need to install it manually.

   bash scripts/install_openbabel.sh

Finally) Install the Current Repository

After setting up your environment, proceed with the repository installation. Run the following command in your terminal or VS Code terminal to install the Python package located in the current directory:

pip install .

This installs the package (along with any dependencies specified) into the Python environment inside your Docker or dev container setup.

Preparing inference data

We provide the inference data from all baseline models and the proposed model. pdb_dict contains the information of the receptor for each index.

bash scripts/download_data.sh

Note that

• the outputs of the models "ar", "ligan", and "targetdiff" were derived from those of https://github.com/guanjq/targetdiff
• AR(auto-regressive generative model) failed to generate SDF file for 5, 6, 72 target.
• the outputs of the model "pocket2mol" were produced using the code provided in https://github.com/pengxingang/Pocket2Mol
• FLAG(Fragment based model) failed to generate SDF file for 14, 77 target.

Sampling

To reproduce the contents of test_outputs/*/p2mrl_SDF, run the following. Prerequisites: test_outputs/*/receptor.pdb.

bash scripts/p2m_rl_inference_on_test_set.sh

Evaluation

To reproduce the contents of test_cache/test_*_outputs, run the following. Prerequisites: test_outputs/*/*_SDF and test_cache/ref_docking_score.pkl.

scripts/compute_metrics.sh

Note that there are unhandled sources of randomness regarding the computation of docking scores etc., which might result in slightly different outcomes from those in the paper.

To obtain the summary statistics that appear in the paper, run summarize_results.ipynb.

Datasets used in the paper

You don't need to download the datasets to reproduce the results of the paper.

Please refer to README.md in the data folder.