目的/意义研究一种均方根误差最小准则的偏最小二乘筛选中药药效物质方法,以便全面地观察和分析中药的作用机理.方法/过程 以均方根误差(root mean square error,RMSE)最小为主要准则,通过偏最小二乘法获得特征的变量投影重要性指标(variable importance in the projection,VIP)值,再以VIP值的大小对特征重要性排序,最后通过偏最小回归法与前向搜索法,以RMSE最小、交叉性验证结果最好为标准,确定特征子集.采用大承气汤配比治疗急性胰腺炎实验数据,以及麻杏石甘汤治咳、平喘、退热实验数据进行验证.结果/结论该方法能得到回归性能最好时的最小RMSE和药效物质子集.VIP值大于1的特征是相对重要的,VIP值小于1的特征也可能对模型性能有影响.
A Partial Least Squares Screening Method for Pharmacological Substances of Traditional Chinese Medicine Based on the Criteri-on of Minimum Root Mean Square Error
Purpose/Significance To study a partial least squares(PLS)screening method for traditional Chinese medicine(TCM)pharmacological substances based on the minimum root mean square error(RMSE)criterion,in order to observe and analyze the action mecha-nism of traditional Chinese medicine comprehensively.Method/Process The main objective is to obtain the variable importance in the projec-tion(VIP)values of features using the PLS method with the minimum RMSE as the main criteria.Then the importance of features is ranked based on the VIP values.Finally,by using PLS regression and forward search methods,the feature subset is determined based on the criteria of the minimum RMSE and obtaining the best cross validation results.Experimental data on the treatment of acute pancreatitis with the combina-tion of DaChengQi Tang and MaXingShiGan Tang for cough,asthma,and fever are verified.Result/Conclusion This method can obtain the minimum residual and subset of pharmacological substances with the best regression performance.Features with a VIP value greater than 1 are relatively important,and features with a VIP value less than 1 may also have an impact on the model performance.
partial least squares(PLS)variable importance in the projection(VIP)root mean square error(RMSE)feature se-lectionsubstance screening
万方数据
维普
