摘要
采用支持向量回归(SVR)算法~([1,2]),构建了偶氮染料的亲和力(A)与被筛选出的四个分子参数(n:染料中共轭键的长度;a,c,λ_3:笛卡尔坐标中的空间参数)之间的非线性关系。留一法交叉验证的结果表明该模型具有很好的泛化能力,其均方根误差(RMSE)为1.38。本工作所用的方法为该类染料提供了亲和力(A)预报的有效方法。
Support Vector Regression with Gaussian kernel function was applied to construct the relationship between the affinities of azo dyestuff molecules and their structural parameters including the length of the conjugated chain of the dye molecule(n) and Steric parameters(a,c,λ_3) derived from Cartesian coordinates.The result of LOOCV test suggest that the QSPR model has good generalization ability with the RMSE is 1.38.Therefore,the method can be used to predict the affinity in computer aided molecular design of azo dyestuff.
引文
[1]Vapnik Vladimir N,The Nature of statistical Learning Theory.Berlin,Springer,1995
[2]Wencong Lu,XiaoboJi,Minjie Li,Liang Liu.Advances in Manufacturing.2013,1(2):151-159