ChironRNA: Steric Clashes Resolution in RNA Structures via E(3)-Equivariant Diffusion.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Experimental RNA structures often contain geometric defects due to limited data resolution.
• These defects, including steric clashes and missing atoms, hinder accurate biological mechanism interpretation.
• Traditional RNA structure refinement methods struggle with the complexity of these defects.

Purpose of the Study:

• To develop an advanced computational method for refining RNA structures.
• To address and resolve physically implausible geometries in determined RNA models.
• To improve the geometric accuracy of RNA structures for functional studies.

Main Methods:

• ChironRNA, an all-atom diffusion model utilizing E(3)-equivariant graph neural networks (EGNN).
• A hierarchical approach combining all-atom and coarse-grained diffusion models.
• A two-stage pipeline involving training and generation for step-by-step atom regeneration.

Main Results:

• ChironRNA achieved an 80% reduction in steric clashes for over 80% of the test set.
• Demonstrated high performance on RNA structures under 200 nucleotides, with >80% clash reduction and 100% atom reconstruction.
• Successfully resolved steric clashes and reconstructed missing atoms with high precision.

Conclusions:

• ChironRNA offers a robust solution for RNA structure refinement, overcoming limitations of traditional methods.
• The model effectively corrects geometric defects, improving the reliability of RNA structural data.
• Enhanced RNA structural accuracy facilitates deeper understanding of RNA biological functions.