Assessing Global-Local Secondary Structure Fingerprints to Classify RNA Sequences With Deep Learning.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Non-coding RNAs (ncRNAs) are crucial for biological processes and diseases.
• RNA structure is vital for function, often conserved across species.
• Predicting ncRNA function relies on secondary structure, but RNA can adopt multiple structures.

Purpose of the Study:

• To address limitations of using single RNA structures for function prediction.
• To develop a novel method for capturing comprehensive RNA structural information.
• To improve the accuracy of ncRNA function classification.

Main Methods:

• Proposed secondary structure fingerprints: RNA-As-Graphs (RAG) for high-level representation and free energy motifs for local structures.
• Developed a deep learning architecture incorporating these fingerprints.
• Evaluated performance using k-mers, specifically 6-mers.

Main Results:

• The proposed global-local structure fingerprints capture diverse RNA folding patterns.
• Combining fingerprints with 6-mers achieved high classification performance.
• Achieved 91.04% accuracy, 91.10% precision, and 91.00% recall.

Conclusions:

• Secondary structure fingerprints offer a more informative representation of ncRNA structure than single structures.
• This approach enhances the accuracy of ncRNA function prediction.
• The developed deep learning model demonstrates significant potential for ncRNA research.