A Multimodal Framework for Improving in Silico Drug Repositioning With the Prior Knowledge From Knowledge Graphs

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning is crucial for discovering new disease treatments.
• Large biological datasets aid drug repositioning, but efficient utilization is challenging.
• Integrating diverse data sources is key to improving in silico drug repositioning.

Purpose of the Study:

• To develop a novel multimodal framework, GraphPK, for enhanced in silico drug repositioning.
• To leverage prior knowledge from drug knowledge graphs and biological data.
• To improve the prediction of drug-disease associations.

Main Methods:

• Constructed a knowledge graph integrating drugs, diseases, and their associations.
• Applied knowledge graph embedding to extract drug and disease prior knowledge.
• Utilized known drug-disease associations and biological domain features (chemical structures, semantic similarity).
• Designed a multimodal neural network to combine diverse features for prediction.

Main Results:

• GraphPK significantly outperforms state-of-the-art methods in predicting drug-disease associations.
• Ablation studies confirm the critical role of knowledge graph prior knowledge in predictive power and robustness.
• Case studies suggest GraphPK's potential for practical application in drug discovery.

Conclusions:

• GraphPK offers a robust and effective approach for in silico drug repositioning.
• Integrating prior knowledge from knowledge graphs substantially enhances prediction accuracy.
• The multimodal framework demonstrates potential for real-world drug discovery applications.