摘要
采用分子描述符计算软件PCLIENT获得123个醇类有机小分子化合物的1 666个理化性质参数,通过相关性分析与逐步线性回归筛选,最终获得14个分子描述符。基于保留的14个关键理化性质,分别以多元线性回归(MLR)、偏最小二乘回归(PLS)与支持向量回归(SVR)构建醇类化合物对欧洲林蛙蝌蚪毒性的QSAR模型。结果表明:3种模型的独立预测决定系数Q2从初始的-163.350、-0.019、0.686分别提升到0.860、0.903与0.936,剔除无关描述符能显著提升模型的预测精度;基于SVR的训练拟合精度和独立预测精度均较好,表明其泛化能力强,鲁棒性好;SVR模型独立测试集预测值和真实值比较结果证明最终筛选出的14个描述符具有较好的显著性,模型具有较好的稳健性。本方法在有毒化合物等QSAR研究领域有较广泛应用前景。
The molecular descriptor calculation software PCLIENT was used to obtain 1 666 physicochemical parameters from 123 alcohol organic small molecule compounds. Through the correlation analysis and stepwise linear regression screening,only 14 molecular descriptors were finally reserved. Based on these 14 critical physicochemical properties,QSAR models were established to predict the toxicity of alcohol compounds to tadpole of European Rana by multiple linear regression( MLR),partial least squares regression( PLS) and support vector regression( SVR),respectively. The results showed that the independent predictive determination coefficients( Q2) for 3 models were raised from the initial values of-163. 350,-0. 019 and 0. 686 to0. 860,0. 903 and 0. 936,respectively. The removal of irrelevant descriptors could significantly improve the prediction accuracy of models. Based on SVR,the fitting accuracy and prediction accuracy both were better,which indicated that the model had better generalization ability and robustness. The comparative results of predicted value and true value from SVR model showed that the screened 14 descriptors had better significance,and the model had better robustness. The method had a wide application prospect in QSAR research of toxic compounds.
出处
《山东农业科学》
2017年第2期67-71,共5页
Shandong Agricultural Sciences
基金
湖南省自然科学基金项目(2016JJ6060)
关键词
醇类化合物
欧洲林蛙蝌蚪
描述符筛选
支持向量回归
定量构效关系
Alcohol compound
Tadpole of European Rana
Descriptor screening
Support vector regression
Quantitative structure-activity relationship