Reinforcement Learning-Based Molecular Generation for RET Inhibitor Discovery

Overview

This repository contains the code and configuration files for reinforcement learning-based molecular generation targeting RET (Rearranged during Transfection) kinase inhibitors.

Publication: AI-Human Collaborative Design and Structure-Activity Relationship Study of Quinoxaline Derivatives for Targeting RET Alterations

Authors: Vinay Pogaku; Surendra Kumar; Ji-hoon Oh; Mi Hye Kim; Kyoung-jin Min; Sung Min Ahn; Han Joo Maeng; Mi-hyun Kim*

Affiliation: Gachon Institute of Pharmaceutical Science and Department of Pharmacy, College of Pharmacy, Gachon University, 191 Hambakmoe-ro, Yeonsu-gu, Incheon, Republic of Korea

*Corresponding author: kmh0515@gachon.ac.kr

---

Requirements

• REINVENT v3.2 - Deep reinforcement learning framework for de novo drug design
• Python 3.7+ with dependencies specified in REINVENT
• Schrödinger Software Suite (2020-4 or later) - Requires valid license (for molecular docking/scoring)
• OpenEye Scientific Software - Requires valid license (for cheminformatics operations)

---

Installation

Installing REINVENT v3.2

Follow the official installation instructions from the REINVENT GitHub repository:

git clone https://github.com/MolecularAI/Reinvent.git
cd Reinvent
conda env create -f reinvent_shared.yml
conda activate reinvent-shared.v3.2

For detailed setup instructions, please refer to the REINVENT documentation.

---

Molecular Generation Workflow

1. Configuration Files

The repository includes multiple JSON configuration files located in the RL_JSON_FILE/ folder. Each configuration file defines:

• Scoring functions for molecular optimization
• Prior and agent models
• Reinforcement learning parameters
• Diversity filters and constraints

2. Running RL Jobs

To execute a reinforcement learning job:

cd RL_JSON_FILE/Job01
python path_to_reinvent_dir/input.py RL_config.json

⚠️ Important: Update all file paths in the JSON configuration files before execution to match your local directory structure.

3. Inception File

The inception_file/ contains known RET-specific ligands used to guide the reinforcement learning process. This file:

• Provides initial chemical diversity
• Biases generation toward RET-relevant chemical space
• Can be directly referenced in REINVENT configuration files

Usage: Specify the path to the inception file in your *_config.json under the inception parameters.

---

Repository Structure

.
├── RL_JSON_FILE/          # Configuration files for RL jobs
│   ├── Job01/
│   ├── Job02/
│   └── ...
├── inception_file/        # RET-specific ligands for guided generation
└── readme.md

---

Workflow Summary

1. Configure the reinforcement learning parameters in JSON files
2. Specify scoring functions (docking, property prediction, etc.)
3. Run REINVENT with the provided configurations
4. Generate novel molecular structures optimized for RET binding
5. Evaluate generated molecules using your own analysis pipeline

---

Citation

If you use this code or methodology in your research, please cite:

Vinay Pogaku; Surendra Kumar; Ji-hoon Oh; Mi Hye Kim; Kyoung-jin Min; Sung Min Ahn; Han Joo Maeng; Mi-hyun Kim
"AI-Human Collaborative Design and Structure-Activity Relationship Study of Quinoxaline Derivatives for Targeting RET Alterations"
[Journal details to be added upon publication]

---

Contact

For questions, issues, or collaboration inquiries:

• Mi-Hyun Kim: kmh0515@gachon.ac.kr
• Surendra Kumar: surendramph@gmail.com

---