基于完备集预测芳香类化合物13C核磁共振波谱  

Prediction of 13C NMR Chemical Shifts of Aromatic Compounds Based on Complete Sets of Descriptors

作　　者：史珍珠 禹新良[1,2] SHI Zhen-zhu;YU Xin-liang(Hunan Provincial Key Laboratory of Environmental Catalysis & Waste Regeneration, Hunan Institute of Engineering, Xiangtan 411104, China;State Key Laboratory of Chemo/Biosensing and Chemometrics, Hunan University, Changsha 410082, China)

机构地区：[1]湖南工程学院环境催化与废弃物再生化湖南省重点实验室,湘潭411104 [2]湖南大学化学生物传感与计量学国家重点实验室,长沙410082

出　　处：《湖南工程学院学报（自然科学版）》2019年第3期69-73,共5页Journal of Hunan Institute of Engineering(Natural Science Edition)

基　　金：湖南省自然科学基金(12JJ6011);化学生物传感与计量学国家重点实验室(湖南大学)开放课题(2016013);环境催化与废弃物再生化湖南省重点实验室(湖南工程学院)开放课题(2018KF11)

摘　　要：在建立分子定量结构-性能关系(QSPR)模型过程中,需要挑选分子结构参数子集.但目前还没有统一的参数挑选方法,参数集所用的分子参数可多可少,带有主观性.参数完备集具备哲学之美,不多一个元素,也不少一个元素.本文基于完备集建立61种芳香类化合物695个13C核磁共振(NMR)化学位移(δC)QSPR模型.完备集分子参数基于PBE1PBE/6-311G(2d,2p)量子化学方法计算得到.采用Duplex算法对数据集进行划分,并用支持向量机(SVM)结合粒子群优化算法(PSO)建立13CNMR化学位移的QSPR模型.所建的2个SVM模型对整个数据集预测的均方根误差(rms)均为2.4ppm,小于广义回归神经网络(GRNN)模型预测结果;此结果与文献报道值相比也是精确的.结果表明,应用完备集建立13CNMR化学位移SVM预测模型是成功的;且为QSPR建模提供了新的参数集挑选方法.Choosing the best set of descriptors is a key step for developing quantitative structure-property relationship (QSPR) models. However, nowadays there is no general method to reveal descriptor’s importance. Thus choosing the best set of descriptors is subjective and different descriptor set may be obtained for the same case. A complete set of descriptors means that the set is perfectness since there is no any element redundant or need to be added. Here we report the application of complete sets of descriptors calculated with PBE1PBE/6-311G(2d,2p) approach to develop QSPR models for 695 13 C NMR chemical shifts (δ C parameters) of carbon atoms in 61 aromatic compounds. Duplex algorithm is used to split the data sets. Two QSPR models for δ C parameters are developed with support vector machine (SVM) algorithm, by applying the particle swarm optimization (PSO) technique to optimize SVM parameters. Both of the two SVM models have root mean square ( rms ) errors of 2.4 ppm for the total set of 695 δ C parameters, which are less than the errors from two general regression neural network (GRNN) models. Compared with previous QSPR models for 13 C NMR chemical shifts, the prediction results are accurate, which suggest that applying complete sets of descriptors for SVM models is successful. This study provides a new selection method of descriptor set in QSPR modeling.

关 键 词：芳香类化合物 完备集 NMR化学位移 构效关系 支持向量机 粒子群优化算法 

分 类 号：O657.61[理学—分析化学] O642.1[理学—化学]