VSMVI方法预测取代芳烃类化合物淡水发光菌的毒性

摘要

选取2组取代芳烃类共53个化合物为目标化合物,以Dragon描述符表征其分子结构信息,采用基于变量相互作用的变量筛选方法(VSMVI)对训练集化合物进行变量寻优.以Dragon描述符为自变量,以化合物的毒性值作为因变量,然后根据化合物样本数和奥卡姆剃刀原则,运用通用变量筛选与建模程序(GVSMP)中线性建模的最优建模法,建立了化合物分子结构与其毒性间的相互关系模型——定量结构-毒性关系(QSTR)模型.结果表明,模型的复相关系数R2和交互验证相关系数Q2LOO均在0.90以上,所构建的模型均具有较高的拟合能力和预测能力.此外,通过模型校验方法对模型进行校验,也显示出模型具有较好的外部估计能力.%The Dragon descriptors are used to characterize the molecular structure of two groups consisting of 53 substituted benzene compounds.All compounds and training sets compounds were optimized respectively by the variable selection method based on variable interaction.On the principle of the number sample compounds and Occam'razor compounds,using Dragon descriptor as independent variable,the toxic value of the compound was used as dependent variable.Then optimal linear modeling construction act,which is from the General Variable Select and Modeling Program (GVSMP),is used to establish the relationship between toxicant quality (dependent variable) and thecompound of inter-quantitative structure-toxicity relationship.The determination coefficient (R2) and the leave-one-out cross-validated correlation coefficient (Q2) are all above 0.90.The results show that the models have high internal predictive abilities and fitting abilities.In addition,the models were verified by the relevant verification method,and have good external estimation ability.