Multi-Knowledge Graph and Multi-View Entity Feature Learning for Predicting Drug-Related Side Effects.

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery
• Artificial Intelligence in Healthcare

Background:

• Predicting drug side effects is critical for patient safety and efficient drug development.
• Existing graph-based methods often fail to leverage the full potential of diverse knowledge graph topologies and semantics.
• There is a need for advanced computational models that can integrate multi-view information for accurate drug-side effect association prediction.

Purpose of the Study:

• To propose MVDSA, a novel multi-view model for drug-side effect association prediction.
• To effectively integrate multiple relationship semantics, local graph topologies, and multi-view entity pair features.
• To improve the accuracy and reliability of computational drug side effect prediction.

Main Methods:

• Construction of two knowledge graphs based on drug/side effect similarities and known associations.
• Development of a space-sensitive learning strategy with relation-gated semantic encoders for adaptive feature learning.
• Implementation of a connection-sensitive tail entity attention mechanism and a knowledge graph-level attention mechanism for feature fusion.
• Utilization of a multi-view enhanced multi-layer perceptron (MLP) for final association prediction.

Main Results:

• MVDSA significantly outperformed 10 state-of-the-art methods in drug-side effect association prediction.
• Ablation studies confirmed the effectiveness of the proposed components, including attention mechanisms and multi-view learning.
• Case studies demonstrated MVDSA's ability to identify potential adverse drug reactions for specific medications.

Conclusions:

• MVDSA offers a robust and effective approach for predicting drug-side effect associations by integrating diverse knowledge graph information.
• The model's ability to capture complex relationships and multi-view features enhances its predictive power.
• MVDSA holds promise for advancing computational drug safety assessments and aiding pharmaceutical research.