摘要
目的将神经网络模型与传统中药药性理论(四性、五味、归经)相结合来预测分析中药肾毒性,方法通过文献检索具有肾毒性证据的中药,并将《中华本草(精选本)》中去除上述肾毒性中药的其他中药作为非肾毒性中药纳入数据。以《中华本草》为标准,确定每味中药的四性、五味和归经归属,分别进行肾毒性/非肾毒性中药与其四性、五味、归经因素的相关性检验,筛选出相关性变量因素,用于构建神经网络模型(Neural Networks Model,NNM)。同时,绘制模型的"受试者工作特征曲线"(Receiver Operator Characteristic Curve,ROC曲线),并计算曲线下面积(Area Under the Curve,AUC),用于评估模型的预测能力。结果肾毒性/非肾毒性中药与四性、五味归属具有相关性(P<0.05),与归经归属无相关性(P>0.05)。NNM结果显示,热性、辛味、温性和苦味是影响中药肾毒性的前4位重要因素,热性排在重要性第1位,模型ROC曲线的AUC计算结果为0.739。结论将传统中药理论与现代数理统计方法相结合建立的中药肾毒性神经网络模型具有一定的预测性,该建模方法可为中药肾毒性及中药毒理学研究提供一定的参考。
Objective To study the correlation of nephrotoxicity of CHMs with the drug-property (four properties,five flavors,and channel tropism) for the establishment of prediction models of nephrotoxicity based on Neural Networks Model (NNM),and to evaluate the models.Methods A comprehensive literature search was performed for CHMs with nephrotoxicity.The CHMs with nephrotoxicity were collected from Chinese Materia Medica·Featured,and the rest were assigned to be CHMs without nephrotoxicity.The data information was collected from Chinese Materia Medica,covered the drug-property (containing the four properties,five flavors and channel tropism) of the included herbs.The correlations of nephrotoxic or non-nephrotoxic CHMs with the four properties,five flavors and channel tropism were analyzed separately.Then the related variables were input into the database.The NNM was constructed using the variables screened.The ROC curves of the model were drawn and the AUCs were calculated;the accuracy of the prediction model was then evaluated.Results CHMs with or without nephrotoxicity were correlated with the four properties and five flavors (P < 0.05).NNM was established.The four variables,including hot,pungent,warm and bitter were selected by NNM.And the top one of predictive variables in NNM was hot.The AUC of ROC curves of the models was 0.739.Conclusion Combining Traditional Chinese Herbal Medicine theory with modern mathematical statistics method,the NNM had been established for predicting nephrotoxicity of CHMs and showed a certain accuracy.The modern mathematics method of data statistics might be a promising method for studying nephrotoxicity and toxicology of CHMs.
作者
刘红杰
董含秋
陈亮
吴倩
LIU Hongjie;DONG Hanqiu;CHEN Liang;WU Qian(School of Traditional Chinese Medicine,Jinan University,Guangzhou 510632 Guangdong,China)
出处
《中药新药与临床药理》
CAS
CSCD
北大核心
2019年第5期622-629,共8页
Traditional Chinese Drug Research and Clinical Pharmacology
基金
国家自然科学基金青年科学基金项目(30901936)
关键词
中药
药性
四性
五味
归经
肾毒性预测
神经网络模型
Chinese Herbal Medicines
drug-property
four properties
five flavors
channel tropism
prediction of nephrotoxicity
Neural Networks Model
