Multi-branch Convolutional Neural Network for Identification of Small Non-coding RNA genomic loci

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic regions encoding small RNA genes possess distinct sequence, secondary structure, and evolutionary conservation patterns.
• Convolutional Neural Networks (CNNs) are powerful tools for data classification based on learned patterns.

Purpose of the Study:

• To introduce MuStARD, a CNN-based application for identifying genomic regions encoding small RNA genes.
• To demonstrate MuStARD's capability to learn patterns from user-defined genomic regions and scan large genomic areas for novel, similar regions.
• To showcase MuStARD's generic applicability across different classes of human small RNA genomic loci without requiring domain-specific knowledge.

Main Methods:

• Application of Convolutional Neural Networks (CNNs) to identify genomic regions.
• Development of MuStARD, a CNN application for pattern learning and genomic region scanning.
• Automated feature and background selection processes within the MuStARD model.
• Demonstration of inter-species identification of functional elements using trained models.

Main Results:

• MuStARD successfully learns patterns associated with small RNA genes.
• The method demonstrates generic applicability across various human small RNA genomic loci.
• MuStARD accurately predicts mouse small RNAs (pre-miRNAs and snoRNAs) using human-trained models.
• Successful application in filtering small RNA-Seq datasets for novel small RNA loci identification in human, mouse, and fly.

Conclusions:

• MuStARD is a versatile and effective tool for identifying small RNA genomic loci.
• The method facilitates both intra- and inter-species identification of functional genomic elements.
• MuStARD's automated processes and ease of deployment make it extendable to diverse genomic classification tasks.