Logratio变换与PSO-BP神经网络在多目标混料设计药物处方配比优化中的应用  

The Application of Logratio Transform and PSO-BP Neural Network in the Optimization of Multi-objective Mixture Design Drug Prescription Ratio

作　　者：李一汀 乔宇超 王旭春 任家辉 崔宇 赵执扬 刘静 赵瑞青 仇丽霞 Li Yiting;Qiao Yuchao;Wang Xuchun(Department of Health Statistics,Public Health of School,Shanxi Medical University,Taiyuan 030001)

机构地区：[1]山西医科大学公共卫生学院卫生统计学教研室,030001

出　　处：《中国卫生统计》2025年第1期44-49,共6页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(81973155)。

摘　　要：目的研究Logratio变换、PSO-BP神经网络及改进非劣分类遗传算法(NSGA-Ⅱ)在药物处方配比优化中的应用,为药物混料设计的优化问题提供科学、合理的方法。方法针对复方甘草微乳混料试验数据,先对数据进行Logratio变换,之后以微乳粒径和有效成分皮肤滞留量两个评价指标为输出构建PSO-BP神经网络模型,再以PSO-BP为适应度函数采用NSGA-Ⅱ进行多目标寻优,最后将本文优化方案与原文优化方案进行比较。结果以粒径和有效成分皮肤滞留量作为输出的PSO-BP神经网络拟合模型的决定系数分别为R^(2)=0.97298和R^(2)=0.96334,且与原文使用的Scheffe多项式模型相比拟合效果更好。采用NSGA-Ⅱ优化目标函数所得3、4、6、7、10、11等方案的复方甘草微乳制备效果均优于原文方案,其中3号方案与原文方案相比,微乳粒径减小了3.02 nm,有效成分皮肤滞留量提高了18.31μg。结论将Logratio变换和PSO-BP神经网络结合应用于混料设计所得试验数据的模型构建中,并采用NSGA-Ⅱ获得最佳的药物处方配比,理论是可行且合理的。Objective In order to provide a scientific and reasonable method for the optimization of drug mix design,the application of PSO-BP neural network modeling after Logratio transformation and nondominated sorting genetic algorithmⅡ(NSGA-Ⅱ)optimization in the optimization of drug prescription ratio of multi-objective mix design was explored.Methods Based on the analysis of experimental data in literature,after Logratio transformation of experimental data of compound glycyrrhiza microemulsion,PSO-BP neural network model was constructed by taking particle size and skin retention of active components as evaluation indexes,and then NSGA-Ⅱwas adopted for multi-objective optimization of the network.Finally,the optimization scheme in this paper was compared with that in the original paper.Results The fitting effect of PSO-BP neural network using particle size and active component skin retention as output is R^(2)=0.97298 and R^(2)=0.96334,respectively,indicating that the fitting effect of PSO-BP is better,and the fitting effect is improved compared with the Scheffe polynomial model used in the original paper.In this paper,PSO-BP was used to construct the model,and NSGA-Ⅱscheme 3、4、6、7、10、11 etc.were superior to the original scheme.Compared with the original scheme,the microemulsion particle size was reduced by 3.02nm and the skin retention of the active ingredient was increased by 18.31μg.Conclusion In theory,it is feasible and reasonable to use Logratio transformation and PSO-BP neural network in the model construction of mixed data and NSGA-Ⅱalgorithm to obtain the optimal ratio of drug prescription.

关 键 词：混料设计 Logratio变换 PSO-BP神经网络 改进非劣分类遗传算法 

分 类 号：R195.1[医药卫生—卫生统计学]