LncRNA-miRNA interaction prediction based on multi-source heterogeneous graph neural network and multi-level...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying long non-coding RNA (lncRNA)-microRNA (miRNA) interactions is crucial for understanding biological processes and disease mechanisms.
• Traditional experimental methods for lncRNA-miRNA interaction discovery are inefficient.
• Existing graph neural network methods often oversimplify intermediate node selection and neglect information importance in heterogeneous networks.

Purpose of the Study:

• To develop a novel model, LMI-MM, for predicting potential lncRNA-miRNA interactions.
• To improve upon existing graph neural network approaches by incorporating multi-source heterogeneous networks and a multi-level attention mechanism.
• To enhance the accuracy and efficiency of lncRNA-miRNA interaction prediction.

Main Methods:

• Constructed homogeneous and multi-source heterogeneous networks including lncRNAs, miRNAs, diseases, drugs, and mRNAs.
• Employed graph neural networks and graph representation learning for feature extraction.
• Introduced multi-level attention modules for adaptive information aggregation and weighting.

Main Results:

• LMI-MM demonstrated superior performance compared to existing models, evidenced by high AUC and AUPR values.
• The model effectively captured the varying importance of different information types through differentiated weighting.
• Case studies confirmed LMI-MM's capability in identifying potential lncRNA-miRNA interactions.

Conclusions:

• LMI-MM offers a powerful and effective approach for predicting lncRNA-miRNA interactions.
• The model's multi-source heterogeneous graph and multi-level attention mechanisms significantly improve prediction accuracy.
• This method aids in advancing disease diagnosis and treatment strategies through a deeper understanding of lncRNA-miRNA regulatory networks.