Different Graph-Level Attention Based on Multi-Scale for Predicting lncRNA-Disease Associations.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) play critical roles in disease pathogenesis.
• Accurate prediction of lncRNA-disease associations aids biological research and therapeutic target identification.
• Existing prediction methods often neglect multi-scale graph information.

Purpose of the Study:

• To develop an advanced graph neural network framework for predicting lncRNA-disease associations.
• To integrate global structural, local subgraph, and multi-scale graph information for enhanced prediction accuracy.
• To introduce GLALDA, a novel prediction framework utilizing multi-scale graph-level attention.

Main Methods:

• Designed a graph neural network (GNN) prediction framework named GLALDA.
• Employed a dual-level attention mechanism to integrate global and local graph information.
• Incorporated edge features and cross-attention for multi-scale feature fusion.

Main Results:

• GLALDA achieved high performance with AUC of 0.949 and AUPR of 0.947 on public datasets.
• Outperformed six state-of-the-art methods in lncRNA-disease association prediction.
• Case studies on three cancers validated GLALDA's ability to identify potential disease-related lncRNAs.

Conclusions:

• GLALDA effectively predicts lncRNA-disease associations by integrating multi-scale graph information.
• The framework offers a promising tool for advancing biological research and clinical applications.
• GLALDA enhances efficiency and precision in identifying novel therapeutic targets.