gCoSRNA: Generalizable Coaxial Stacking Prediction for RNA Junctions Using Secondary Structure.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Coaxial stacking of RNA stems is crucial for RNA tertiary structure and spatial organization.
• Accurate prediction of coaxial stacking is essential for RNA 3D structure modeling.
• Existing computational tools struggle with complex RNA junctions.

Purpose of the Study:

• To develop a generalizable computational framework, gCoSRNA, for predicting RNA coaxial stacking configurations.
• To overcome limitations of existing methods in handling variable or complex RNA junctions.

Main Methods:

• gCoSRNA decomposes multi-way RNA junctions into pseudo two-way junctions.
• A unified Random Forest classifier predicts stacking probabilities for these pairs.
• Global configurations are inferred by integrating pairwise predictions, avoiding junction type classification.

Main Results:

• gCoSRNA demonstrates high accuracy (mean ~0.89) across junctions with two to seven branches.
• The framework outperforms existing junction-specific methods on independent test sets (CASP15/16, RNA-Puzzles).
• Achieves consistent accuracy regardless of junction complexity or topology.

Conclusions:

• gCoSRNA provides a robust and generalizable approach for predicting RNA coaxial stacking.
• The framework effectively captures higher-order structural features in RNA.
• gCoSRNA has significant potential for integration into RNA tertiary structure prediction pipelines.