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

Designing RNA sequences that reliably fold into specific secondary structures is essential for understanding their biological functions but remains a challenging computational problem. We propose CocoRNA, a cooperative multi-agent reinforcement learning framework for RNA inverse design. 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. As a result, CocoRNA learns high-quality RNA design policies that generalize effectively to unseen structures without additional training. Experiments on the Rfam dataset demonstrate that 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 and broad potential of CocoRNA for complex design tasks.

RNA molecules play essential roles in biology, but designing RNA sequences that reliably fold into a desired structure remains extremely difficult because the number of possible sequences grows exponentially with sequence length. Existing AI and optimization methods often struggle with this enormous search space, becoming slow or trapped in poor solutions. We introduce CocoRNA, a cooperative multi-agent reinforcement learning framework that treats RNA design as a collective problem solved by multiple AI agents working together. Instead of having a single model design an entire RNA sequence alone, CocoRNA divides the task into smaller subproblems and assigns them to different agents, while a centralized training mechanism coordinates their decisions using global structural information. This cooperative strategy improves exploration of the RNA design space and enables the system to learn effective design policies that generalize to unseen RNA structures without retraining. Experiments show that CocoRNA substantially outperforms existing state-of-the-art methods in both accuracy and efficiency.