A Hyperbolic Discrete Diffusion 3D RNA Inverse Folding Model for Functional RNA Design.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Generative design of functional RNAs offers significant potential for RNA-based biotechnologies and biomedical applications.
• RNA inverse folding is a key strategy for designing RNA sequences with specific 3D structures.
• 3D RNA inverse folding is challenging due to limited experimental 3D structure data and the complexity of RNA 3D conformations.

Purpose of the Study:

• To introduce RIdiffusion, a hyperbolic denoising diffusion generative model for 3D RNA inverse folding.
• To address the challenges of limited data and complex structures in 3D RNA design.
• To develop a powerful tool for generating functional RNAs.

Main Methods:

• Developed RIdiffusion, a discrete diffusion model operating in hyperbolic space.
• Integrated geometric features and topological properties of RNA 3D structures into the hyperbolic embedding.
• Utilized limited training samples for efficient recovery of nucleotide distributions.

Main Results:

• RIdiffusion demonstrated superior performance compared to baseline generative models in RNA inverse folding tasks.
• The model consistently outperformed other methods across various datasets and data-splitting strategies.
• Achieved efficient recovery of nucleotide distributions for targeted 3D RNA structures.

Conclusions:

• RIdiffusion is an effective tool for generative RNA inverse folding, particularly in data-scarce environments.
• The model's hyperbolic approach enhances performance by capturing geometric and topological RNA features.
• RIdiffusion holds promise for advancing RNA-based biotechnologies and biomedical applications through improved functional RNA design.