Fusing graph transformer with multi-aggregate GCN for enhanced drug-disease associations prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Identifying drug-disease associations is crucial for drug discovery and safety.
• Computational methods are vital but face challenges with heterogeneous network data.
• Accurate prediction aids in discovering new drug indications and reducing adverse reactions.

Purpose of the Study:

• To develop an advanced computational framework for predicting drug-disease associations.
• To effectively integrate heterogeneous network data for improved prediction accuracy.
• To enhance drug repositioning strategies and drug safety research.

Main Methods:

• Proposed WMAGT framework fusing Graph Transformer Networks and multi-aggregate graph convolutional networks.
• Constructed heterogeneous information graphs integrating drug-drug, drug-disease, and disease-disease networks.
• Employed graph transformer with self-attention and neural collaborative filtering for feature representation.

Main Results:

• WMAGT demonstrated robust and effective performance in predicting drug-disease associations.
• The framework accurately models local and global node interactions within the heterogeneous graph.
• Experimental results showed superior performance compared to existing state-of-the-art methods.

Conclusions:

• WMAGT significantly outperforms current methods in drug-disease association prediction.
• The proposed model is beneficial for advancing drug repositioning and ensuring drug safety.
• The study validates the effectiveness and robustness of the WMAGT framework through rigorous testing.