摘要
基于遗传算法(GA)的因子筛选和支持向量机(SVM)的非线性回归,提出了1种改进的有机物定量结构-性质相关(QSPR)建模方法——遗传-支持向量机(GA-SVM),并将其用于38种食品工业常用有机物正辛醇-水分配系数(Kow)的QSPR建模.结果显示,QSPR模型选取了分子量、Hansen极性、沸点、含氧率和含氢率5种参数;模型的预测值与实测值间的误差平方和(SSE)、均方差(RMSE)和决定系数(R2)分别为0.048、0.036和0.999,表明模型具有较强的预测能力;同时,交叉验证的结果(SSE=0.295,RMSE=0.089,R2=0.995)也表明,模型具有良好的稳健性,因此,GA-SVM算法适用于对有机物正辛醇-水分配系数的QSPR建模.此外,将基于GA-SVM的QSPR模型分别与基于遗传-径向基神经网络(GA-RBFNN)和基于线性算法的模型进行了比较,结果表明,应用GA-SVM建立的QSPR模型无论从稳健性还是预测能力上都优于应用其它2种算法建立的模型,因此,GA-SVM算法比GA-RBFNN和线性算法更适合于对有机物正辛醇-水分配系数进行QSPR建模.
A modified method to develop quantitative structure-property relationship （QSPR） models of organic compounds was proposed based on genetic algorithm （GA） and support vector machine （SVM） （GA-SVM）. GA was used to perform the variable selection, and SVM was used to construct QSPR models. GA-SVM was applied to develop the QSPR models for n-octanol-water partition coefficients （ Kow ） of 38 typical organic compounds in food industry. 5 descriptors （molecular weights, Hansen polarity, boiling point, percent oxygen and percent hydrogen） were selected in the QSPR model. The coefficient of multiple determination （R^2）, the sum of squares due to error （SSE） and the root mean squared error （RMSE） values between the measured values and predicted values of the model developed by GA-SVM are 0.999, 0.048 and 0.036, respectively, indicating good predictive capability for lgKow values of these organic compounds. Based on leave-one-out cross validation, the QSPR model constructed by GA-SVM showed good robustness （SSE = 0.295, RMSE = 0.089, R^2 = 0.995）. Moreover, the models developed by GA-SVM were compared with the models constructed by genetic algorithm-radial basis function neural network （GARBFNN） and linear method. The models constructed by GA-SVM show the optimal predictive capability and robustness in the comparison, which illustrates GA-SVM is the optimal method for developing QSPR models for lgKow values of these organic compounds.
出处
《环境科学》
EI
CAS
CSCD
北大核心
2008年第1期212-218,共7页
Environmental Science
基金
国家重点基础研究发展规划(973)项目(2003CB415204)
