Drug-disease association prediction using semantic graph and function similarity representation learning over...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Discovering new indications for existing drugs is crucial for drug development.
• Current methods often overlook higher-order connectivity in biological networks, limiting prediction accuracy.
• Integrating diverse biological knowledge into network models is essential for improved drug-disease association prediction.

Purpose of the Study:

• To propose SFRLDDA, a novel computational model for predicting drug-disease associations (DDAs).
• To leverage semantic graph and function similarity representation learning within a heterogeneous information network (HIN).
• To enhance the accuracy and comprehensiveness of identifying potential DDAs.

Main Methods:

• Constructed a heterogeneous information network (HIN) integrating drug-disease, drug-protein, and protein-disease associations with biological knowledge.
• Applied representation learning strategies to semantic graph and function similarity graphs for drug and disease feature extraction.
• Utilized a Random Forest classifier to predict potential drug-disease associations (DDAs).

Main Results:

• SFRLDDA demonstrated superior performance compared to state-of-the-art models on three benchmark datasets.
• The model achieved high accuracy in predicting drug-disease associations.
• Case studies confirmed the model's ability to precisely predict DDAs.

Conclusions:

• SFRLDDA effectively predicts drug-disease associations by integrating semantic and functional similarities within a HIN.
• The model offers a comprehensive approach to discovering new drug indications.
• This computational strategy holds significant promise for advancing drug research and development.