MMnc: multi-modal interpretable representation for non-coding RNA classification and class annotation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-coding RNAs (ncRNAs) play crucial biological roles and are implicated in diseases.
• Characterizing novel ncRNAs is vital for identifying biomarkers and therapeutic targets.
• The complexity of ncRNA data poses a significant challenge for accurate classification.

Purpose of the Study:

• To introduce MMnc, an interpretable deep-learning model for classifying non-coding RNAs into functional groups.
• To leverage multi-modal data integration for robust ncRNA characterization.
• To provide insights into ncRNA patterns through interpretable attention mechanisms.

Main Methods:

• MMnc utilizes an attention-based multi-modal data integration approach.
• It incorporates diverse data sources including sequence, secondary structure, and expression data.
• The model is designed to handle missing data sources in certain samples.

Main Results:

• MMnc demonstrates high classification accuracy across various non-coding RNA classes.
• Its modular architecture supports multiple data modalities, outperforming tools with limited input.
• The method is resilient to missing data, maximizing the utility of available information.
• Attention scores provide interpretable insights into ncRNA class patterns.

Conclusions:

• MMnc offers an effective and interpretable solution for non-coding RNA classification.
• The approach facilitates deeper understanding of ncRNA functions and potential applications.
• This tool can advance research in ncRNA-based biomarkers and therapeutics.