RNA Design via Conditioned Flow Matching and Finite-Policy Reinforcement Learning

2026-09-29

AI Summary

This study addresses the limitation of existing RNA sequence design methods that fail to explicitly model the co-variation and selection mechanisms inherent in natural evolution. To this end, this work proposes a two-stage generative framework that combines Dirichlet-structured conditional flow matching with reinforcement learning to simulate co-variation, while introducing thermodynamic feedback to optimize sequence selection. By constructing a constrained policy space that preserves base-pairing properties, the proposed approach establishes an evolution-driven paradigm for RNA generation. Evaluated on the Rfam-27 benchmark, the method achieves a Pass@1 accuracy of 85.19%, demonstrating significant improvements in both the thermodynamic stability and robustness of the generated sequences.

Abstract

RNA design aims to identify sequences that fold into specified secondary structures. Existing methods formulate the task as target-specific search or conditional generation. However, natural RNA evolution proceeds through sequence variation and selection, with compensatory substitutions, whereas these methods do not explicitly model this process. To address this limitation, we propose a two-stage framework comprising RNA Inverse-Folding Flow (RNA-IFlow) and RNA-IFlow-RL. RNA-IFlow uses structure-conditioned Dirichlet Flow Matching to model coordinated variation across the sequence, while RNA-IFlow-RL maps the learned flow to a pairing-preserving finite policy and refines it with thermodynamic feedback. Our framework achieves leading performance on multiple benchmarks, reaching 85.19% Pass@1 on Rfam-27. Further analyses reveal thermodynamic gains, policy dynamics, and robustness across settings. Our work couples coordinated variation with thermodynamic selection, offering a novel paradigm for RNA design.

Problem

• RNA design
• inverse folding
• secondary structure
• sequence evolution
• compensatory substitutions

Innovation

• Dirichlet Flow Matching
• Finite-Policy Reinforcement Learning
• RNA Inverse Folding
• Thermodynamic Feedback
• Compensatory Substitutions