CocoRNA: Collective RNA Design with Cooperative Multi-agent Reinforcement Learning

Forum topic · 小凯 · 2026-10-20

Summary

CocoRNA is a cooperative multi-agent reinforcement learning framework for RNA inverse design, proposed by Tianmeng Hu, Biao Luo, and Ke Li (arXiv:2610.23820). The method addresses the challenge of designing RNA sequences that reliably fold into target secondary structures. CocoRNA decomposes the design task into smaller sub-problems, each solved collaboratively by multiple agents, which lowers problem complexity and improves exploration of design policies. During training, a centralized critic leverages global structural information to guide the agents, enabling joint optimization of their design strategies. The learned policies generalize effectively to unseen RNA structures without additional training. Experiments on the Rfam dataset show that CocoRNA substantially outperforms state-of-the-art methods in both success rate and design speed. Further experiments on other biological sequence design tasks demonstrate the framework's effectiveness and its broad potential for complex design problems.

Paper Overview

Field: Machine Learning
Authors: Tianmeng Hu, Biao Luo, Ke Li
Published: 2026-10-16
arXiv: 2610.23820

Abstract

Designing RNA sequences that reliably fold into specific secondary structures is essential for understanding their biological functions but remains a challenging computational problem. The authors propose CocoRNA, a cooperative multi-agent reinforcement learning framework for RNA inverse design.

Approach

CocoRNA simplifies the design task by decomposing it into smaller sub-problems, each solved collaboratively by multiple agents. This approach reduces the complexity of the problem and improves the exploration of design policies. During training, a centralized critic uses global structural information to guide the agents, enabling them to jointly optimize their design strategies.

Results

CocoRNA learns high-quality RNA design policies that generalize effectively to unseen structures without additional training.
On the Rfam dataset, CocoRNA substantially outperforms state-of-the-art methods in both success rate and design speed.
Further experiments on other biological sequence design tasks highlight the effectiveness of CocoRNA and its broad potential for complex design tasks.