球坐标变换和主成分二次推断函数在缓控释制剂混料处方优化中的应用研究

发布时间：2018-09-04 21:04

【摘要】：目的:将球坐标变换、主成分二次推断函数建模方法及改进非劣分类遗传算法应用于混料设计缓控释制剂处方优化,解决缓控释制剂混料数据的定和约束、共线性及评价指标间的重复测量的问题获得药物累积释放度达到最佳释放目标时的最优处方配比,评价整套优化方案的效果,为混料设计缓控释制剂的建模与优化提供一套合理、可行的方案。方法:对处方配比含零与不含零两个实例进行探索性研究,以各时点累积释放度为因变量,以各混料组分为自变量,利用球坐标变换消除自变量间的定和约束,主成分方法解决变量间的共线性问题,运用二次推断函数方法,分别基于可交换相关矩阵、一阶自相关矩阵、无结构相关矩阵建立模型,根据评价指标AIC,BIC选择较优模型,运用改进非劣分类遗传算法(NSGA-Ⅱ)进行优化,最后球坐标反变换获得处方配比,与原文献的优化结果比较。结果:在尼莫地平骨架片处方配比不含零的处方优化中,混料组分球坐标变换后,采用主成分分析提取了7个主成分,可解释原信息的99.98%,运用二次推断函数建模,以可交换相关矩阵建立的模型有统计学意义(Q=4.35,P=0.82),效果较好(AIC=20.3451,BIC=25.4575)。采用改进非劣分类遗传算法优化得到Pareto最优解集:所有目标均在处方筛选范围之内,其中有多个方案12h累积释放度(Q12)在99%以上,能达到较好释放。当HPMC、乳糖、海藻酸钠的比例分别为:0.2649、0.6464、0.0887时,Q12可达到99.60%,3h、6h、9h的累积释放度分别为:21.21%、50.66%,77.60%,均在处方筛选范围内。原文献通过构建Scheffé多项式模型,用等高线图法从图形中主观挑选了5个解构成解方案集,其中有两个方案9小时累积释放度不在规定解范围内。所有方案中12小时累积释放度没有99%以上的解。当HPMC、乳糖、海藻酸钠的比例分别为:0.3458、0.4715、0.1627时,Q12最高才达到98.65%,比本文优化结果低了0.95%。在甲硝唑缓释片处方配比含零的处方优化中,混料组分球坐标变换后,采用主成分分析提取了10个主成分,可解释原信息的99.97%,运用二次推断函数建模,以可交换相关矩阵建立的模型有统计学意义(Q=4.67,P=0.95),效果较好(AIC=26.6661,BIC=37.6192)。采用改进非劣分类遗传算法优化得到Pareto最优解集:多个方案1~4天预测累积释放度与释放目标的差距不超过2%,5天累积释放度均在90%以上。当甲硝唑、PCL、HPMC、GMS比例分别为0.044、0.817、0.115、0.023时,1~5天累积释放度分别为40.32%、55.31%、70.9%、85.08%、92.13%。原文通过建立Scheffé多项式模型寻找到7个最优方案,当第1天和第4天预测累积释放度都达到释放标准时,其他时点预测累积释放度与目标相差很大,不能同时接近最佳释放目标。结论:采用球坐标变换消除混料组分的定和约束,主成分二次推断函数解决共线性及累积释放度间的相关性,将二者结合应用于缓控释制剂混料数据建模,并采用NSGA-Ⅱ多目标遗传算法进行组分配比优化,再进行反变换获得符合需要的最佳配方配比,这一整套方案应用于混料设计缓控释制剂的优化研究,是可行且合理的。对于有效解决混料设计缓控释制剂的多目标优化问题,有一定应用价值。
[Abstract]:OBJECTIVE: To apply spherical coordinate transformation, principal component quadratic inference function modeling method and improved non-inferior classification genetic algorithm to formulation optimization of sustained and controlled release preparations for mixture design, solve the problem of data constraints, collinearity and repeated measurements among evaluation indexes, and obtain the optimal cumulative drug release rate. METHODS: Two cases of zero and zero-free formulations were studied. The cumulative release rate at each time point was taken as a dependent variable, and the components of each mixture as an independent variable. Scalar transformation eliminates the definite sum constraints between independent variables and principal component analysis solves the problem of collinearity among variables. Using quadratic inference function method, the model is established based on commutative correlation matrix, first-order autocorrelation matrix and unstructured correlation matrix respectively. According to the evaluation index AIC and BIC, the better model is selected and the improved non-inferior classification genetic algorithm (NSGA-BIC) is used. Results: In the formulation optimization of nimodipine skeleton tablets without zero, seven principal components were extracted by principal component analysis after spherical coordinate transformation, which could explain 99.98% of the original information. The model based on interchangeable correlation matrix was statistically significant (Q = 4.35, P = 0.82), and the effect was better (AIC = 20.3451, BIC = 25.4575). The Pareto optimal solution set was obtained by using improved non-inferior classification genetic algorithm. All the targets were within the prescription selection range, and the cumulative release rate (Q12) of several schemes was above 99% in 12 hours, which could achieve better release. When the ratios of HPMC, lactose and sodium alginate were 0.2649, 0.6464 and 0.0887 respectively, the cumulative release rates of Q12 were 99.60%, 21.21%, 50.66% and 77.60% for 3, 6, and 9 hours respectively, which were all within the prescription selection range. The cumulative release rate of Q12 reached 98.65% at the ratios of HPMC, lactose and sodium alginate 0.3458, 0.4715 and 0.1627 respectively, which was 0.95% lower than the optimized results in this paper. In the optimization, 10 principal components were extracted by principal component analysis after spherical coordinate transformation, which could explain 99.97% of the original information. The model established by quadratic inference function was statistically significant (Q = 4.67, P = 0.95) and the effect was better (AIC = 26.6661, BIC = 37.6192). The optimization was carried out by improved non-inferior classification genetic algorithm. The Pareto optimal solution set was obtained: the difference between the predicted cumulative release rate and the release target in 1-4 days was less than 2%, and the cumulative release rate in 5 days was more than 90%. When the ratios of metronidazole, PCL, HPMC and GMS were 0.044, 0.817, 0.115, 0.023, the cumulative release rates in 1-5 days were 40.32%, 55.31%, 70.9%, 85.08% and 92.13% respectively. When the cumulative release rate reached the release standard on the first day and the fourth day, the cumulative release rate at other time points was very different from the target and could not approach the optimal release target at the same time. And the correlation between cumulative release rate, the combination of the two methods was applied to modeling the data of sustained and controlled release formulations, and NSGA-II multi-objective genetic algorithm was used to optimize the composition ratio, and then inverse transformation was carried out to obtain the optimal formulation ratio. This whole set of schemes was applied to the optimization research of sustained and controlled release formulations. It is of certain application value to solve the multi-objective optimization problem of slow release controlled release mixture design effectively.
【学位授予单位】：山西医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R943

【参考文献】