Drug repurposing is a highly effective strategy in drug development that identifies new therapeutic applications for existing drugs, offering accelerated timelines and reduced risks compared to traditional de novo approaches. With the rapid growth of large-scale biomedical data, new computational methods have been developed to predict relationships between drugs and diseases, thereby facilitating drug repurposing efforts. Nevertheless, current methods often fail to capture heterogeneous information in biomedical networks, including diverse node types and multityped edges, which limits their prediction accuracy. To address this challenge, we propose MEGCAM, a model that effectively extracts heterogeneous information from the constructed biomedical information network through the Meta-Graph technique. And a path selection strategy based on a causal attention mechanism has been designed to provide effective guidance for information aggregation. Experimental results demonstrate that MEGCAM performs competitively with state-of-the-art models, showing strong robustness and generalizability across two independent data sets. Furthermore, a case study on Alzheimer’s disease demonstrates MEGCAM’s practical utility and reliability. Overall, MEGCAM shows great potential to accelerate therapeutic discovery by enhancing prediction accuracy and model interpretability.
Liu et al. (Thu,) studied this question.