NTFold: Structure-Sensing Nucleotide Attention Learning for RNA Secondary Structure Prediction

Area of Science:

• Computational Biology
• Molecular Biology
• Bioinformatics

Background:

• Determining RNA secondary structures is essential but challenging.
• Experimental methods are costly and time-consuming.
• Accurate prediction methods are needed for large-scale applications.

Purpose of the Study:

• Introduce NTFold, a deep learning framework for accurate RNA secondary structure prediction.
• Improve upon existing computational methods for RNA structure determination.
• Enable efficient and scalable RNA structural sensing.

Main Methods:

• Developed NTFold, integrating a Nucleotide Attention Module (NAM) and a Structural Refinement Module (SRM).
• NAM models dependencies among nucleotides for fine-grained sequence correlations.
• SRM refines correlation maps, preserving spatial information and ensuring structural consistency.

Main Results:

• NTFold achieves high-precision contact maps for reliable RNA secondary structure reconstruction.
• Demonstrated superior performance compared to existing deep learning-based predictors.
• Effectively learns both local and global nucleotide interactions.

Conclusions:

• NTFold offers a novel approach for RNA secondary structure prediction.
• The framework integrates attention-based correlation modeling with structure-sensing refinement.
• Provides an efficient and scalable solution for RNA structural sensing.