RNA design improves with flow matching and reinforcement learning
RNA Design via Conditioned Flow Matching and Finite-Policy Reinforcement Learning
Machine Learning
Summary
Designing RNA means finding sequences that fold into a target shape, but current methods don’t mimic how nature evolves RNA by changing sequences step-by-step. The authors created a two-step process that first models how sequence changes work together, then refines this approach using feedback about the RNA’s stability. Their method performs very well on standard tests and shows it can produce better-folded RNA sequences more reliably. This approach combines coordinated sequence changes with energy feedback to design RNA more like evolution does.
What this means in practice
- •For biotech engineers: Create RNA sequences that reliably fold into desired shapes for synthetic biology and therapeutic development.$Commercial implications: This method enables production of custom RNA molecules for drug development and synthetic biology, potentially selling design tools or RNA products.
- •For bioinformatics developers: Develop improved RNA design software using coordinated sequence variations and energy feedback for better folding predictions.
Authors
Zefeng Lin, Xianyong Fang, Tianfan Fu, Xiaohua Xu
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.