HGANMDA: A Heterogeneous Graph Adversarial Network for Multimodal Microbe-Drug Association Prediction.

Area of Science:

• Biomedical informatics
• Computational biology
• Network pharmacology

Background:

• Accurate microbe-drug association (MDA) prediction is essential for antimicrobial therapy and drug repositioning.
• Experimental validation is costly and time-consuming.
• Existing models struggle with heterogeneous and multiscale biomedical network interactions.

Purpose of the Study:

• To develop an advanced computational model for predicting microbe-drug associations.
• To address limitations of current methods in capturing complex biological network structures.

Main Methods:

• Developed HGANMDA, a heterogeneous graph adversarial network.
• Integrated multimodal biological data into a unified heterogeneous graph.
• Employed a multichannel structural encoder with attention-based aggregation.
• Introduced adversarial embedding regularization for enhanced robustness and feature separability.

Main Results:

• HGANMDA consistently outperformed state-of-the-art baseline models across multiple metrics on three benchmark datasets.
• Demonstrated superior performance in predicting microbe-drug associations.
• Validated the effectiveness of the proposed heterogeneous graph learning approach.

Conclusions:

• Adversarially regularized heterogeneous graph learning shows significant potential for advancing antimicrobial research.
• HGANMDA offers a robust and accurate method for predicting microbe-drug associations.
• The findings support the use of advanced network learning techniques in drug discovery and development.