基于PSO-ν-SVM算法的1-芳基-四氢异喹啉类化合物抗HIV活性的QSAR建模研究  

QSAR Studies on 1-aryl-tetrahydroisoquinoline Analogs as Active Anti-HIV Agents Based on PSO-ν-SVM

作　　者：吕新旗[1] 蒲红玉[1] 张运陶[1] 

机构地区：[1]西华师范大学应用化学研究所,四川南充637009

出　　处：《西华师范大学学报（自然科学版）》2010年第3期302-308,共7页Journal of China West Normal University(Natural Sciences)

摘　　要：以E-Dragon软件计算的RDF,WHIM,Topological,2D-autocorrelation,Geometrical,3D-MoRSE和GETA-WAY等7种QSAR建模常用的分子描述符为结构参数,分别在PSO算法筛选变量的基础上,再以ν-SVM算法对36种1-芳基-四氢异喹啉类化合物的抗HIV活性药物进行定量构效关系(QSAR)研究.7种分子描述符建立的PSO-ν-SVM模型中,以RDF描述符建立的模型最佳.该模型对训练集和预测集计算结果的平均绝对误差MAE分别为0.0028和0.0630,决定系数R2分别为0.998和0.956;而文献[1]3个模型训练、预测结果的MAE分别为0.0612,0.5486,0.0557,0.5676和0.0665,0.5658,训练结果的R2分别为0.9493,0.9533和0.9286,预测结果的R2则均小于0;研究表明,该QSAR模型明显优于文献[1]的3个模型.The study for the drug quantitative structure-activity relationship（QSAR） of 36 kinds of 1-aryl-Tetrahydro-isoquinoline analogs of anti-HIV activity has been carried on,taking the commonly used seven kinds of QSAR modeling of molecular descriptors,RDF,WHIM,Topological,2D-autocorrelation,Geometrical,3D-MoRSE and GETAWAY calculated from the E-Dragon software as the structural parameters respectively,based on variable selection with ν-SVM algorithm.Among the PSO-ν-SVM models established from the seven kinds of molecular descriptors,the RDF model descriptor is the best.The average absolute errors MAE of the model in the training set and test set are respectively 0.0028 and 0.0630,coefficients of determination with R2 are 0.998 and 0.956,respectively;and the MAE results of training set and test set of the reference [1] 3 models are 0.0612,0.5486,0.0557,0.5676 and 0.0665,0.5658,the results of training set with R2 are 0.9493,0.9533 and 0.9286,respectively,the results of test set with R2 are less than 0.The research illuminates that the QSAR model is superior to the three models of the reference[1].

关 键 词：微粒群 ν-支持向量机 定量构效关系 抗艾滋病毒 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]