RNAbpFlow: base pair-augmented SE(3) flow matching for conditional RNA 3D structure generation

2026-06-30 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Engineering & Applied Sciences · Depth: Expert, extended

Summary

RNAbpFlow is a novel sequence- and base pair-conditioned SE(3)-equivariant flow-matching model designed for generating all-atom RNA 3D structural ensembles. It operates end-to-end using a nucleobase center representation, eliminating the need for evolutionary or homologous structural templates. The model demonstrated superior performance over the MD simulation-based RNAJP, achieving a mean lDDT of 0.66 versus 0.59 and a mean TM-score of 0.38 versus 0.32 in topology sampling. On CASP15 targets, RNAbpFlow reached an average TM-score of 0.48 with native base pairs, a 20% improvement over predicted inputs. For CASP16 targets (≤200 nucleotides), it surpassed top automated servers like AF3-server and Yang-Server for hard targets with shallow MSA, and outperformed AF3, NuFold, trRosettaRNA2, and DRfold2 in MSA-free settings. Data augmentation and fine-tuning further enhanced its accuracy.

Key takeaway

For research scientists and machine learning engineers focused on biomolecular structure prediction, RNAbpFlow offers a powerful, MSA-free approach for RNA 3D structure generation. You should prioritize integrating accurate base-pairing information and consider data augmentation via cross-distillation to enhance model performance. This method provides a strong foundation for generating diverse RNA conformational ensembles, particularly for targets where evolutionary data is scarce.

Key insights

RNAbpFlow accurately generates all-atom RNA 3D structures by conditioning SE(3) flow matching on sequence and base pair information.

Principles

• Base-pairing information is critical for RNA 3D generative modeling.
• Data augmentation significantly boosts model accuracy.
• Input base pair quality directly impacts structural fidelity.

Method

RNAbpFlow employs an SE(3)-equivariant flow-matching model with a nucleobase center representation. It integrates sequence and three base pair annotation methods as conditions, optimizing rotatable bond angles and using base pair-centric auxiliary losses.

In practice

• Combine multiple base pair annotation methods for robust conditioning.
• Utilize nucleobase center representation for end-to-end all-atom generation.
• Incorporate base pair-centric auxiliary losses to improve conditional fidelity.

Topics

• RNA 3D Structure Prediction
• SE(3) Flow Matching
• Base Pair Conditioning
• Deep Generative Models
• Biomolecular Modeling
• CASP Challenges

Code references

• Bhattacharya-Lab/RNAbpFlow
• marcellszi/rna3db
• rcsb/RNAView
• Shujun-He/RibonanzaNet
• jaswindersingh2/SPOT-RNA

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• RNA design across eras: from covariance models to modern generative AI — RNA's functional versatility relies on 3D folding, with generative AI building on foundational probabilistic models for structure prediction.
• Predicting RNA 3D structure and conformers using a pre-trained secondary structure model and structure-aware attention — trRosettaRNA2 accurately predicts RNA 3D structures and conformers by integrating pre-trained secondary structure priors and SS-aware attention.
• Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language — ConfSeq enables large language models to process 3D molecular structures by converting them into discrete token sequences.
• Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching — Atom ordering in neural representations can encode chemical knowledge, improving retrosynthesis model efficiency and accuracy.
• Programmable RNA translation through deep learning-driven IRES discovery and de novo generation — An AI framework unifies IRES identification, optimization, and de novo generation for programmable RNA translation.
• Carbon, open source DNA model, 250x faster than Evo2-7B and runs on llama.cpp — Applying LLM architectures to DNA sequences enables highly efficient genomic analysis and structure prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.