MncR: Late Integration Machine Learning Model for Classification of ncRNA Classes Using Sequence and Structural...

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Non-coding RNAs (ncRNAs) are crucial for cellular functions and regulation.
• Accurate ncRNA classification is vital for understanding cellular mechanisms and discovering biomarkers.
• Existing classification methods face challenges due to ncRNA heterogeneity.

Purpose of the Study:

• To develop an improved machine learning model for classifying diverse non-coding RNA classes.
• To integrate primary sequence and secondary structure information for enhanced ncRNA classification accuracy.
• To evaluate the performance of the new model against existing state-of-the-art tools.

Main Methods:

• Utilized primary sequence and graph-encoded secondary structure data from RNAcentral.
• Developed and trained machine learning models, including neural networks, for ncRNA classification.
• Focused on six major ncRNA classes: lncRNA, rRNA, tRNA, miRNA, snRNA, and snoRNA.

Main Results:

• The MncR classifier achieved an overall accuracy exceeding 97% through late integration of sequence and structure features.
• No significant improvement in accuracy was observed with more granular subclassification.
• MncR demonstrated a minimal 0.5% accuracy increase over ncRDense on overlapping ncRNA classes.
• The model successfully predicts long non-coding RNA classes up to 12,000 nucleotides.

Conclusions:

• MncR offers superior accuracy and broader applicability for ncRNA classification compared to current tools.
• The model's ability to handle long ncRNAs expands its utility in genomic research.
• The findings underscore the importance of integrating diverse data types for robust ncRNA analysis.