Embedding multilayer RNA networks for lncRNA-miRNA interaction prediction via LMI-MHGAT

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying long noncoding RNA-microRNA interactions (LMIs) is vital for understanding gene regulation in health and disease.
• Current computational methods struggle with integrating multimodal data and handling imbalanced biological network data.

Purpose of the Study:

• To develop a robust deep learning framework for predicting LMIs.
• To overcome limitations of existing methods in data integration and class imbalance.

Main Methods:

• Developed LMI-MHGAT, a Multilayer Heterogeneous Graph Attention network framework.
• Integrated RNA sequences, expression profiles, and molecular interactions into a unified graph.
• Employed a graph attention mechanism for dynamic weighting of relational information.

Main Results:

• LMI-MHGAT significantly outperforms 14 existing methods on human LMI prediction.
• Demonstrated exceptional robustness with a 1:60 positive-to-negative ratio.
• Achieved state-of-the-art performance on rat and plant datasets, validating generalization capabilities.

Conclusions:

• LMI-MHGAT offers a powerful and robust solution for LMI prediction.
• The framework effectively addresses data integration and utilization challenges.
• The tool is publicly available for research use.