Computational biomedicine is a critical field for improving our understanding of biological systems and drug discovery and development. Existing computational models address several problems on this domain such as drug-drug interaction (DDI) prediction, drug-target interaction prediction, drug-disease association prediction, and drug-target-disease association prediction. Heterogeneous biomedical graphs are gaining attention to address different problems because of their ability to model distinct biomedical entities including drug and protein and ability to capture complex relationships between these distinct entities.

In this study, by leveraging the power of heterogeneous graphs to address key challenges in computational biomedicine, we develop novel heterogeneous graph learning models for drug-target-disease association prediction and DDI prediction. Our models capture intricate drug-target-disease and drug-drug relationships, enabling accurate predictions. In addition, our models outperform state-of-the-art methods on real-world datasets, demonstrating the potential of heterogeneous graphs for improving our understanding of complex biomedical interactions and advancing the development of safer and more effective therapeutics.