基于符号图卷积网络的药物互作用关系预测

Drug-Drug Interactions Prediction Based on Signed Graph Convolutional Network

药物相互作用(DDI)是指两种或两种以上药物在药理行为方面的相互影响.大多数现有计算方法都是针对传统的二元预测而设计的,即确定DDI是否发生.然而,确定DDI是增强的还是抑制的,对于治疗和护理病人更具意义.受图神经网络启发,尝试利用符号图卷积网络来预测DDI的符号.将该问题转换为一个嵌入问题,然后通过对数几率回归获得DDI类型.结果表明,符号图卷积网络在该问题上表现良好,具有可行性.

Drug-drug interactions(DDIs)refer to the interaction of two or more drugs in pharmacological behavior.Most of existing calculation meth⁃ods are designed for traditional binary predictions,i.e.predicting the possibility of DDI for a drugs pair.However,it is more meaningful for the treatment and nursing of patients to determine whether DDI is enhanced(positive)or depressive(negative).predicting the possibility of DDI for a drugs pair.Inspired by the successful applications of graph neural networks,we employ the Signed Graph Convolution Network(SGCN)to predict the sign of DDI.Nodes embedding results are obtained via SGCN,and then DDIs are predicted by logistic regression on embedding pairs.As a result,in experiments SGCN shows its feasibility on DDIs prediction problem.