HGDD: A Drug-Disease High-Order Association Information Extraction Method for Drug Repurposing via Hypergraph

Traditional drug research and development (R&D) methods are characterized by high risk and low efficiency. Drug repurposing provides a feasible way for the efficiency and safety of drug R&D. Since high-precision prediction of drug-disease association can help us quickly locate potential treatment options for existing drugs, how to accurately predict drug-disease association becomes the key of drug repurposing. In this paper, we propose a method to extract high-order drug-diseases association information using hypergraph, named Drug-Disease association prediction base on HyperGraph (HGDD). Specifically, HGDD first extracts the network topology information from the drug-disease association network based on the random walk strategy as the initial features of the nodes. Then, HGDD constructs the drug-disease hypergraph network based on the drug-disease association network. Finally, HGDD uses hypergraph neural network (HGNN) to aggregate higher-order information on hypergraph and predict the association between drugs and diseases. Compared with other traditional drug repurposing methods, HGDD shows substantial performance improvement. The area under the precision recall curve (AUPR) index of HGDD is significantly higher than other controlled trials. Case study also shows that HGDD can discover new associations that do not exist in our dataset. These results indicate that HGDD is a reliable method for drug-disease association prediction.