基于人工神经元网络预测药物共晶粒径  

Prediction of Particle Size of Pharmaceutical Cocrystal by Artificial Neural Network

作　　者：范祥元 宋宇 蒋俊龙 陈言庆 王辉 杜陈侠 FAN Xiangyuan;SONG Yu;JIANG Junlong;CHEN Yanqing;WANG Hui;DU Chenxia(Anhui Food and Drug Inspection and Research Institute,Hefei Anhui 230051,China;Lu'an Food and Drug Inspection Center,Lu'an Anhui 237000,China;Anhui Provincial Drug Evaluation and Inspection Center,Hefei Anhui 230051,China;The First Affiliated Hospital of University of Science and Technology of China,Hefei Anhui 230001,China)

机构地区：[1]安徽省食品药品检验研究院,安徽合肥230051 [2]六安市食品药品检验中心,安徽六安237000 [3]安徽省药品审评查验中心,安徽合肥230051 [4]中国科学技术大学附属第一医院,安徽合肥230001

出　　处：《当代化工》2025年第2期316-322,共7页Contemporary Chemical Industry

基　　金：安徽省科技型中小企业技术创新基金项目(项目编号:23B36912630759)。

摘　　要：采用球磨工艺,以茶碱为原料药,以4-氨基苯甲酸为共晶形成剂,以羟丙基甲基纤维素为辅料,制备了茶碱-4-氨基苯甲酸药物共晶,采用响应面模型研究了辅料质量分数、球料质量比和研磨时间对药物粒径的影响,基于人工神经元网络(ANN)提出了一种球磨法制备药物共晶粒径的预测模型。结果表明:辅料质量分数、球料质量比和研磨时间3个主效应是影响药物粒径的显著因素,辅料质量分数与球料质量比、辅料质量分数与研磨时间、球料质量比与研磨时间是影响粒径的显著交互效应,球料质量比的平方和研磨时间的平方是影响显著的二阶效应。随着辅料质量分数、球料质量比和研磨时间水平的增加,药物共晶粒径呈先降低后增加的趋势,各因素采用中等水平时,药物粒径可以低于70μm。最优结构的ANN模型可以准确预测药物共晶粒径,预测值与实验值相关系数R达到0.9896,高于多元线性回归模型的0.8424。Theophylline-4-aminobenzoic acid pharmaceutical cocrystal was prepared using ball milling process with theophylline as raw material,4-aminobenzoic acid as cocrystal forming agent,and hydroxypropyl methyl cellulose as excipient.The effects of excipient mass fraction,ball-to-material mass ratio,and milling time on pharmaceutical cocrystal particle size were studied using response surface modeling.A prediction model for pharmaceutical cocrystal particle size prepared by ball milling method was proposed based on artificial neural network(ANN).The results showed that three main effects of excipient mass fraction,ball-to-material mass ratio,and milling time were significant factors affecting pharmaceutical cocrystal particle size.The interaction effects of excipient dosage and ball-to-material mass ratio,excipient mass fraction and milling time,and ball-to-material mass ratio and milling time were significant,while square of ball-to-material mass ratio and square of milling time were significant second-order effects.With the increase of excipient mass fraction,ball-to-material mass ratio,and milling time,the particle size of pharmaceutical cocrystal showed a trend of first decreasing and then increasing.When using moderate levels of various factors,the particle size of cocrystal could be lower than 70 μm.The ANN model with the optimal structure could accurately predict the particle size of pharmaceutical cocrystal,and the correlation coefficient R between the predicted values and the experimental values reached 0.989 6,which was higher than the 0.842 4 of the multiple linear regression model.

关 键 词：茶碱 4-氨基苯甲酸 共晶 球磨 粒径预测 人工神经元网络 优化 粉体 

分 类 号：TQ460[化学工程—制药化工]