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

Area of Science:

• Computational Biology
• Structural Biology
• Deep Learning

Background:

• Predicting RNA 3D structures is difficult due to RNA's flexibility and limited data.
• Current deep learning methods face challenges with RNA structure prediction.

Purpose of the Study:

• Introduce RNAbpFlow, a new model for generating RNA 3D structural ensembles.
• Enable end-to-end generation of all-atom RNA structures.

Main Methods:

• Developed RNAbpFlow, a sequence- and base pair-conditioned SE(3)-equivariant flow-matching model.
• Utilized a nucleobase center representation for RNA structure generation.
• Avoided explicit or implicit use of evolutionary information or homologous templates.

Main Results:

• RNAbpFlow successfully generates RNA 3D structural ensembles.
• Base-pairing conditioning significantly improved performance in RNA topology sampling.
• Demonstrated broadly generalizable performance improvements over existing methods in large-scale benchmarking.

Conclusions:

• RNAbpFlow offers a powerful new approach for RNA 3D structure prediction.
• The model advances biomolecular modeling by overcoming data limitations.
• This method enhances the accuracy and generalizability of RNA structural modeling.