De novo Drug Design using Reinforcement Learning with Multiple GPT Agents
Xiuyuan Hu, Guoqing Liu, Yang Zhao, Hao Zhang
TL;DR
MolRL-MGPT introduces a multi-agent reinforcement learning framework using GPT-based agents to perform de novo drug design. By treating molecule generation as a cooperative MARL problem with a shared SMILES prior, a molecular memory, and a diversity-enforcing loss, the method achieves strong performance on the GuacaMol benchmark and demonstrates real-world potential by designing SARS-CoV-2 target inhibitors. Ablation studies show the value of agent collaboration and memory in promoting diversity and high scores, while certain tricks like similarity penalization offer limited benefits. Overall, the work suggests that language-model agents can efficiently explore diverse chemical spaces, offering a scalable approach for rapid candidate generation in drug discovery.
Abstract
De novo drug design is a pivotal issue in pharmacology and a new area of focus in AI for science research. A central challenge in this field is to generate molecules with specific properties while also producing a wide range of diverse candidates. Although advanced technologies such as transformer models and reinforcement learning have been applied in drug design, their potential has not been fully realized. Therefore, we propose MolRL-MGPT, a reinforcement learning algorithm with multiple GPT agents for drug molecular generation. To promote molecular diversity, we encourage the agents to collaborate in searching for desirable molecules in diverse directions. Our algorithm has shown promising results on the GuacaMol benchmark and exhibits efficacy in designing inhibitors against SARS-CoV-2 protein targets. The codes are available at: https://github.com/HXYfighter/MolRL-MGPT.