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

Accurate prediction of microbe-drug associations (MDAs) is vital for guiding antimicrobial therapy and accelerating drug repositioning. Although experimental validation remains the gold standard, it is costly and time-consuming. Existing models, often based on similarity fusion or conventional graph neural networks (GNNs), struggle to capture the heterogeneous and multiscale interaction patterns of biomedical networks. We present HGANMDA, a heterogeneous graph adversarial network for MDA prediction. The framework integrates multimodal biological information into a unified heterogeneous graph, employs a multichannel structural encoder with attention-based aggregation to capture local and global patterns, and introduces adversarial embedding regularization to enhance robustness and feature separability. Experiments on three benchmark data sets show that HGANMDA consistently outperforms state-of-the-art baselines across multiple metrics. These results highlight the potential of adversarially regularized heterogeneous graph learning in supporting antimicrobial research.