Noncoding RNA family classification based on multifeature fusion and convolutional block attention residual network

Area of Science:

• Bioinformatics
• Molecular Biology
• Genomics

Background:

• Noncoding RNAs (ncRNAs) play crucial roles in biological processes, but their classification remains challenging due to sequence length, structural, and functional diversity.
• Accurate characterization requires integrating sequence and structural information, which is difficult to extract effectively.

Purpose of the Study:

• To develop a novel 3D graphical representation method for mining information from RNA secondary structures.
• To propose an advanced ncRNA family classification model, nRMFCA, integrating multifeature fusion and attention mechanisms.

Main Methods:

• A novel 3D graphical representation based on Z-curve and chaos game was developed to convert RNA secondary structures into sequence-based 3D graphical representations.
• The nRMFCA model was designed using multifeature fusion and convolutional block attention residual networks for ncRNA family classification.
• The method's effectiveness was validated on viral sequences, and nRMFCA was compared against existing methods using NCY and nRC datasets.

Main Results:

• The 3D graphical representation method effectively mines underlying base information from RNA secondary structures.
• The nRMFCA model demonstrated superior performance in ncRNA family classification compared to previous methods on both NCY and nRC datasets.
• The integrated approach provides a powerful tool for analyzing and classifying ncRNA families.

Conclusions:

• The novel 3D graphical representation combined with the nRMFCA model significantly enhances ncRNA family classification accuracy.
• This study offers a robust computational tool for advancing ncRNA research and understanding their diverse biological roles.