Predicting human microbe-drug associations based on graph attention network
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
万方数据
维普
目的 采用图注意力网络(graph attention network,GAT)预测人类微生物与药物之间的潜在关联.方法 选取三个常用的微生物-药物关联(microbe-drug associations,MDA)数据集(MDAD、aBiofilm和 Drug Virus),基于数据集中丰富的生物信息构建一个异构网络,并提出一种基于GAT框架预测MDA的模型——GATMDA模型,用于预测微生物与药物间的关联.结果 与现有的 8 种预测方法相比,GATMDA通过三种交叉验证方法在三个数据集上具有较好的预测效果.在 5 折交叉验证的性能评估中,在三个数据集上的受试者工作特征曲线下的面积(area under the curve,AUC)分别为 0.988 6、0.994 1 和 0.983 6,精确率-召回率曲线下的面积(area under the precision-recall curve,AUPR)分别为 0.966 7、0.986 9 和 0.879 5.通过病例研究进一步验证了GATMDA在预测MDA方面的有效性.结论 基于GAT,GATMDA模型可以通过构建的异构网络对微生物-药物进行有效的关联预测.
Objective Graph attention network(GAT)was used to predict the potential association between human microbes and drugs.Methods Three commonly used microbe-drug associations(MDA)datasets including MDAD,aBiofilm and Drug Virus were selected.Based on the rich biological information in the datasets,a heterogeneous network was constructed and a GAT-based framework for predicting MDA called GATMDA was proposed,which was used to predict the association between microbes and drugs.Results Compared with the existing eight prediction methods,GATMDA had better prediction effect on three datasets through three cross-validation methods.During the performance evaluation of 5-fold cross-validation on three datasets,the area under the curve(AUC)were 0.988 6,0.994 1,and 0.983 6,respectively,and the area under the precision-recall curve(AUPR)were 0.966 7,0.986 9 and 0.879 5.The effectiveness of GATMDA in predicting MDA was further validated through case studies.Conclusion Based on GAT,the GATMDA model can effectively predict MDA through the constructed heterogeneous network.
万方数据
维普
