Knowledge Graph Convolutional Network with Heuristic Search for Drug Repositioning.

Area of Science:

• Biomedical informatics
• Computational biology
• Drug discovery

Background:

• Drug repositioning offers a cost-effective and safer alternative to traditional drug development.
• Biomedical knowledge graphs integrate diverse data, presenting opportunities for advanced drug repositioning strategies.
• Existing methods may not fully leverage the semantic and topological information within complex biological knowledge graphs.

Purpose of the Study:

• To propose a novel Knowledge Graph Convolutional Network with Heuristic Search (KGCNH) for predicting drug-disease associations.
• To enhance the utilization of semantic and topological information from biomedical knowledge graphs.
• To improve the accuracy and robustness of drug repositioning predictions.

Main Methods:

• Developed a relation-aware attention mechanism to weigh neighboring entities based on relations.
• Implemented a Gumbel-Softmax-based heuristic search module to explore optimal drug and disease embeddings.
• Utilized neighborhood aggregation for entity embedding and employed feature-based augmented views for model regularization.

Main Results:

• KGCNH demonstrated superior performance compared to existing methods on two benchmark datasets.
• Case studies involving lithium and quetiapine validated KGCNH's ability to identify actual drug-disease associations.
• The model effectively captures both semantic and topological features from biomedical knowledge graphs.

Conclusions:

• KGCNH provides an effective framework for drug repositioning by leveraging biomedical knowledge graphs.
• The proposed heuristic search and attention mechanisms enhance the prediction of drug-disease associations.
• This approach holds significant potential for accelerating the identification of new therapeutic indications for existing drugs.