Learning global dependencies and multi-semantics within heterogeneous graph for predicting disease-related lncRNAs

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Long noncoding RNAs (lncRNAs) are crucial in disease development.
• Predicting lncRNA-disease associations aids in understanding disease mechanisms.
• Existing methods often overlook attribute information within meta-paths.

Purpose of the Study:

• To propose a novel association prediction model, GSMV.
• To deeply integrate global dependencies, meta-path semantics, and multi-view features for lncRNA-disease association prediction.

Main Methods:

• Utilized a self-attention mechanism for global node representations.
• Employed graph neural networks with novel attention mechanisms for meta-path semantics.
• Integrated pairwise multi-view features using dilated convolutions.

Main Results:

• GSMV outperformed seven state-of-the-art methods in lncRNA-disease association prediction.
• Ablation studies confirmed the effectiveness of each proposed module.
• Case studies identified potential lncRNA candidates for three cancers.

Conclusions:

• GSMV effectively predicts lncRNA-disease associations by integrating diverse features.
• The model enhances the understanding of disease pathogenesis.
• GSMV shows promise in discovering novel disease-related lncRNAs.