DeepMiR2GO: Inferring Functions of Human MicroRNAs Using a Deep Multi-Label Classification Model

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial non-coding RNAs regulating cellular processes and implicated in human diseases.
• The functions of many identified miRNAs remain largely uncharacterized.
• Computational approaches are essential for elucidating miRNA functions.

Purpose of the Study:

• To introduce DeepMiR2GO, a novel computational tool for predicting gene ontology (GO) functions of miRNAs.
• To integrate diverse biological networks for enhanced functional prediction.

Main Methods:

• Construction of a global heterogeneous network integrating miRNA co-expression, protein-protein interaction (PPI), and disease phenotype similarity networks.
• Application of graph embedding for learning network representations and topological features.
• Development of a deep multi-label classification network utilizing neuro-symbolic models for GO term annotation.

Main Results:

• DeepMiR2GO effectively predicts gene ontology functions for microRNAs.
• The tool integrates multiple biological data types for comprehensive analysis.
• Performance evaluation shows superior precision, recall, and F-measure compared to existing methods.

Conclusions:

• DeepMiR2GO offers a powerful computational approach for miRNA functional annotation.
• The neuro-symbolic models and integrated network strategy advance miRNA research.
• This tool aids in understanding miRNA roles in health and disease.