基于异构图推断的疾病与药物相关性预测研究

Prediction of Disease and Drug Correlation Based on Heterogeneous Graph Inference

研发药物的过程非常耗时且费用昂贵,以现有药物为基础确定和发展新的治疗效果有利于降低药物的开发成本。而以往的预测方法数据的要求单一,较少考虑到疾病药物相关数据的稀疏性,因此,该篇文章提出了一种基于异构图推断的疾病与药物相关性预测方法(Drug-disease relevant predicted by heterogeneous graph,DDRPGH)。该方法通过将药物相似性和疾病语义相似性与余弦相似性相结合,再通过WKNKN与已知的疾病与药物的关联融合到异构图中,揭示潜在的药物与疾病的关系。在两个数据集的十折交叉验证中,该方法AUC(F:0.923;C:0.939)优于另外三个对比方法,证明了这个方法在疾病与药物的预测方面是可行有效的。

The process of developing drugs is very time-consuming and expensive.Determining and developing new therapeutic effects based on existing drugs is helpful to reduce the cost of drug development.However,the data of previous prediction methods are simple,and the sparsity of disease drug-related data is less considered.Therefore,this paper proposes a prediction method of disease-drug correlation based on heterogeneous graph inference(Drug-disease correlation predicted by heterogeneous graph,DDRPGH).By combining drug similarity and disease semantic similarity with cosine similarity,the method reveals the potential relationship between drugs and diseases by merging WKNKN with known disease and drug associations into heterogeneous maps.In the 10-fold cross validation of two data sets the AUC value of this algorithm is 0.923 and 0.939 which are better than the other three contrast methods.The AUC prove this method is feasible and effective in disease and drug prediction.