DivideFold+: an AI-based tool for RNA secondary structure prediction with subdomains identification and visualization...

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• Predicting RNA secondary structure is crucial for understanding RNA function but challenging for long sequences.
• Existing deep-learning methods struggle with data limitations and accuracy for extended RNA structures.

Purpose of the Study:

• To present DivideFold+, an enhanced deep-learning method for accurate RNA secondary structure prediction.
• To introduce a novel data augmentation strategy tailored for RNA secondary structure prediction.
• To enable segmentation of RNA secondary structures into functional subdomains.

Main Methods:

• Developed DivideFold+, an improved deep-learning approach that fragments long RNAs into shorter, manageable segments.
• Implemented a new, elaborate data augmentation strategy specifically for RNA secondary structure prediction.
• Integrated subdomain segmentation to identify potential functional regions within RNAs.

Main Results:

• The new data augmentation strategy significantly benefits RNA secondary structure prediction accuracy.
• DivideFold+ successfully predicts secondary structures and segments them into biologically relevant subdomains.
• A user-friendly web server facilitates visualization of predicted structures and subdomains.

Conclusions:

• DivideFold+ offers enhanced accuracy and novel subdomain segmentation for RNA secondary structure prediction.
• The developed data augmentation technique is effective for deep-learning-based RNA structure modeling.
• The accessible web server and datasets promote further research in RNA structural biology.