AGRL-DSE: Adaptive Graph Representation Learning on a Heterogeneous Graph for Drug Side Effect Prediction

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Accurate identification of drug side effects is vital for drug development and safety monitoring.
• Graph neural networks (GNNs) show promise but struggle with complex heterogeneous networks and oversmoothing.
• Existing methods often fail to fully capture semantic information in complex network structures.

Purpose of the Study:

• To propose AGRL-DSE, an adaptive graph representation learning framework for enhanced node-feature learning in drug side effect prediction.
• To address limitations in existing GNN-based approaches for heterogeneous networks.
• To improve the accuracy and generalizability of computational drug side effect prediction models.

Main Methods:

• Constructed a heterogeneous graph with intra- and interlayer connections for drugs and side effects.
• Integrated Graph Convolutional Network (GCN), GraphSAGE, and Graph Attention Network (GAT) modules at different graph levels.
• Introduced an adaptive layer attention mechanism for dynamic fusion of multi-level features.

Main Results:

• AGRL-DSE demonstrated superior performance compared to state-of-the-art models in predicting drug side effects.
• Achieved high accuracy in both hot- and cold-start prediction scenarios.
• The adaptive layer attention mechanism effectively fused multi-level features for enhanced prediction.

Conclusions:

• AGRL-DSE offers a robust and generalizable approach for drug side effect prediction.
• The framework's ability to capture complex relationships can significantly impact drug evaluation and development.
• This method holds potential for improving patient safety and optimizing drug discovery processes.