CocoRNA: Collective RNA Design with Cooperative Multi-agent Reinforcement Learning
Keywords: Multi-agent, reinforcement learning, RNA design
TL;DR: We propose CocoRNA, a cooperative multi-agent reinforcement learning framework for RNA secondary structure design, which achieves better solving efficiency and higher success rates.
Abstract: Ribonucleic acid (RNA) plays a crucial role in various biological functions, and designing sequences that reliably fold into specified structures remains a significant challenge in computational biology. Existing methods often struggle with efficiency and scalability, as they require extensive search or optimization to tackle this complex combinatorial problem. In this paper, we propose CocoRNA, a collective RNA design method using cooperative multi-agent reinforcement learning, for the RNA secondary structure design problem. CocoRNA decomposes the RNA design task into multiple sub-tasks, which are assigned to multiple agents to solve collaboratively, alleviating the challenges of the curse of dimensionality as well as the issues of sparse and delayed rewards. By employing a centralized Critic network and leveraging global information during training, we promote cooperation among agents, enabling the distributed policies to cooperatively optimize the joint objective, thereby resulting in a high-quality collective RNA design policy. The trained model is capable of completing RNA secondary structure design with less time and fewer steps, without requiring further training or search on new tasks. We evaluate CocoRNA on the Rfam dataset and the Eterna100 benchmark. Experimental results demonstrate that CocoRNA outperforms existing algorithms in terms of design time and success rate, highlighting its practicality and effectiveness.
Supplementary Material: zip
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11071
Loading