Graph reasoning method enhanced by relational transformers and knowledge distillation for drug-related side effect...

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence

Background:

• Identifying drug side effects is crucial for successful drug development and cost reduction.
• Current methods may not fully leverage complex relationships between drugs and side effects.

Purpose of the Study:

• To propose a novel graph reasoning method, RKDSP, for accurate drug side effect prediction.
• To integrate diverse knowledge sources including semantic relationships, local and global graph knowledge, and node attributes.

Main Methods:

• Constructed drug-side effect heterogeneous graphs incorporating similarity and association connections.
• Employed multiple relational transformers to learn node features from various meta-path perspectives.
• Utilized a knowledge distillation module for local and global knowledge acquisition.
• Developed an adaptive convolutional neural network for encoding drug-side effect pair attributes.

Main Results:

• The RKDSP method demonstrated superior performance compared to existing state-of-the-art prediction approaches.
• The fusion of multiple knowledge sources and relational learning significantly enhanced prediction accuracy.

Conclusions:

• RKDSP offers a powerful new approach for predicting drug side effects.
• This method has the potential to significantly reduce drug development failures and associated costs.