基于Monte Carlo采样模型集群分析对定量结构-色谱保留相关关系(QSRR)常用建模方法的比较研究

A comparative study of QSRR modeling methods based on Monte Carlo sampling model population analysis

本文收集了环烷烃类、环烯烃类、酮类、胺类、醚类、酯类等有机物在固定相角鲨烷和SE-30上的气相色谱保留指数,并采用基于Monte Carlo采样的模型集群分析（Monte Carlo sampling model population analysis,MCS MPA）方法进行了定量结构-色谱保留指数相关关系建模方法的比较研究。对于两种固定相上的有机化合物,分别采用不同的分子描述符予以表征,分子描述符的选择基于统计学与遗传算法。采用的建模方法包括多元线性回归（multivariate linear regression,MLR）、支持向量机回归（support vector machine,SVM）、径向基函数人工神经网络方法（radial basis function artificial neural networks,RBF ANN）,通过所建模型预测了独立外部测试样本的气相色谱保留指数。研究结果表明,对于本文所研究的数据,SVM回归方法的建模效果优于MLR与RBF ANN方法。

The authors performed a comparative study of QSRR modeling methods based on Monte Carlo sampling model population analysis（MCS MPA）. The varieties of GC analytes include naphthenic hydrocarbons, cycloolefins, ethers, amines, ketones, and esters, etc. The retention data of the analytes was collected from two kinds of stationary phases, i.e., squalane and SE-30. Moreover, different molecular descriptors were applied to describe the molecular structures of the analytes on each stationary phase. The molecular descriptors were selected based on statistical methods and genetic algorithm. The modeling tools include multivariate linear regression（MLR）, support vector machine（SVM）, and radial basis function artificial neural networks（RBF ANN）. The retention indices of independent external test samples were predicted by the models. The research shows that the modeling results of SVM were better than those of MLR and RBF ANN.