偏最小二乘在中药药性判别中的应用

目的探讨将偏最小二乘法用于中药药性判别的可行性。方法收集《中华本草》中收录的药性明确、功效主治明确的植物性中药1 725种,利用基于偏最小二乘的两种方法建立药性识别模型,用组内回代、外推预测、交叉验证等方式评价模型。结果基于偏最小二乘法的两种模型的判别效果与原来判定结果符合率分别为92.75%、94.67%,两种方法的外推预测正确率分别为85.51%、89.86%,而重复10次的5折交叉验证正确率分别达到了88.28%、90.46%。结论基于中药的功效主治,偏最小二乘法能够准确判别药性且模型稳定好,能够有效解决多重共线性等问题,为中药药性的有效判别提供了新的思路和方法。

Partial least squares in the discrimination of traditional Chinese herbal medicine property

Objective To investigate the feasibility of partial least squares(PLS) in the discrimination of traditional Chinese herbal medicine property(CHMP).Methods Information of the original effects of 1 725 kinds of Chinese herbal medicine(CHM) was collected from "Chinese Herbal Medicine".Two PLS-based methods were performed to build the discrimination model,which was evaluated by consistent homology rate,extrapolated prediction and 5 fold cross validation.Results Using the two PLS-based methods,consistent homology rates of property with original discriminant results were 92.75% and 94.67%,respectively.Predictive accuracies of the testing set were 85.51% and 89.86%,respectively.10-iteration 5-fold cross validation accuracy were 88.28% and 90.46%,respectively.Conclusion According to original effects of CHM,the PLS-based methods can be useful in discriminating CHMP,with high accuracy and stability,and effectively deal with multicollinearity of multivariate,providing a new idea and method for discrimination of CHMP.