HGBHAN: A Novel Framework for Microbe-Drug Interaction Prediction Using Heterogeneous Graphs and Bi-LSTM With...

Area of Science:

• Biomedical research
• Computational biology
• Drug discovery

Background:

• Predicting microbe-drug associations (MDAs) is vital for drug discovery and clinical interventions.
• Traditional lab methods are costly and slow; computational methods often miss complex network relationships and imbalanced data.
• Existing computational approaches for MDAs prediction face challenges with biological network complexity and data imbalance.

Purpose of the Study:

• To propose HGBHAN, a novel framework for robust microbe-drug association prediction.
• To leverage heterogeneous graphs and Bi-LSTM with hierarchical attention for enhanced MDA prediction.
• To improve the accuracy and scalability of computational methods for predicting microbe-drug associations.

Main Methods:

• Constructing a heterogeneous network integrating microbe/drug similarities and known associations.
• Employing Bi-LSTM modules with hierarchical attention for learning node embeddings.
• Utilizing residual connections to mitigate over-smoothing issues in graph neural networks.

Main Results:

• HGBHAN demonstrated superior performance across multiple evaluation metrics on three public datasets.
• The framework effectively captures multi-level structural and sequential dependencies in biological networks.
• The model achieved higher accuracy in predicting microbe-drug associations compared to existing methods.

Conclusions:

• HGBHAN provides a robust and effective framework for predicting microbe-drug associations.
• The proposed method addresses limitations of previous approaches, particularly regarding network heterogeneity and data imbalance.
• HGBHAN shows significant potential for accelerating drug discovery and optimizing clinical applications.