遗传算法结合人工神经网络模拟药物在超临界流体中溶解度  

Correlation and Prediction of Drug Solubility in Supercritical Fluid Based on Genetic Algorithm and LM-BP Artificial Neural Network

作　　者：何正大[1] 许玫[2] 陈曙[1] 

机构地区：[1]中国药科大学基础部 [2]江苏正大天晴药业股份有限公司,江苏南京210009

出　　处：《中国医药工业杂志》2010年第10期743-747,共5页Chinese Journal of Pharmaceuticals

摘　　要：为更精确地关联预测药物在超临界流体中的溶解度,提出了遗传算法(GA)与LM-反向传播人工神经网络相结合(GA-LM-BPANN)的模型,并设计了该模型的计算过程,讨论了模型参数的设置。用该模型计算了温度(308～348K)和压力(122～355bar)条件下药物(非那吡啶)在超临界CO2中溶解度。结果表明,计算值与实测值的平均相对误差(AARD)为1.53%,测试集的AARD为3.32%。用Bartle半经验方程得到的计算值与实测值的AARD为14.6%。可见,与Bartle半经验方程相比,GA-LM-BPANN模型的关联和预测精度高,关联范围广。A GA-LM-BP ANN based on genetic algorithm（GA） and LM-BP artificial neural network was developed for more accurate correlation and prediction of drug solubility in supercritical fluid.Implementation procedure and parameters setting of this model were introduced in detail.The designed GA-LM-BP ANN with optimized parameters was used to calculate the solubilities of phenazopyridine in supercritical CO2 system at temperature range of 308 348 K and pressure range of 122-355 bar.The calculated results were very close to the experimental solubility values.The average relative deviation（AARD） was 1.53%,and the AARD for the test set was 3.32%.The AARD between calculated results of semi-empirical Bartle model and experimental values was 14.6%.Compared with semi-empirical Bartle model,GA-LM-BP ANN showed less error and wider fitting range.

关 键 词：遗传算法-LM-反向传播人工神经网络 超临界流体 非那吡啶 溶解度 

分 类 号：R94[医药卫生—药剂学]