Path-based Analogical Reasoning for Drug Repurposing
Traditional drug development is costly,inefficient and time-consuming,while drug repurposing methods provide a feasible al-ternative to reduce cost,improve efficiency and shorten time to market.Various knowledge graph-based drug repurposing methods have been proposed with relatively impressive results,but they have limitations such as limiting the scope of the dataset,dealing with simplistic relationships,and neglecting path information between nodes.To compensate for these shortcomings,we propose a drug repurposing approach based on analogical reasoning over the paths between drugs and diseases in knowledge graphs.Initially,multiple biological datasets are integrated to construct a heterogeneous information network.Subsequently,various knowledge graph embedding models(TransE,DistMult,ComplEx,RotatE,and RGCN)are trained to obtain embedding vectors.Then,path ranking algorithm and multilayer perceptron are stacked using AdaBoosted decision stumps to extract original reasoning paths,coupled with analogical reasoning for predictions.Finally,employing traditional performance metrics,embedding evaluations,and reproducibility rates,the TransE model is selected as the prediction model.The proposed approach successfully identifies 10 repurposing candidate drugs and confirms their therapeutic effects through relevant literature,which fully validates its effectiveness.The proposed approach offers a new perspective on drug repurposing by integrating path information,which may benefit other scholars involved in drug repurposing research.
drug repurposingknowledge graph embeddinganalogical reasoningpath ranking algorithmAdaboost
万方数据
维普
