基于分子电性距离矢量预测有机磷杀虫剂的小鼠毒性

Prediction of mouse toxicity of organophosphorous pesticides based on MEDV descriptors

以分子电性距离矢量(MEDV)有效表征22个有机磷杀虫剂(OPs)的分子结构,应用基于预测的变量选择与模型化(VSMP)方法建立OPs分子结构与小鼠半致死量(pLD_(50))的定量结构-毒性相关(QSTR)模型。模型的相关系数(R^2)为0.981 5、估计均方根误差(RMSE)为0.09,留一法(LOO)交叉验证相关系数Q^2为0.967 5,均方根误差(RMSEV)为0.12,说明模型具有良好的稳定性。采用LOO交叉验证、留多法(LMO)交叉验证、自举法和y-随机化对QSTR模型进行内部验证,结果表明模型具有较好的内部稳健性;外部验证表明模型具有较强的预测能力。因此,QSTR模型可预测有机磷杀虫剂类化合物对小鼠的毒性。

Molecular electronegativity distance vector( MEDV) was used to characterize the structures of 22 organophosphorous pesticides,and a quantitative structure-toxicity relationship( QSTR) between the MEDV descriptors and the mouse toxicity was developed using variable selection and modeling based on prediction( VSMP). It was found that the QSTR model with a high correlation coefficient of R^2= 0. 981 5,root mean square error( RMSE) = 0. 09,and correlation coefficient( Q^2) and RMSEV of leave-one-out( LOO) cross-validation is equal to 0. 967 5 and 0. 12,respectively. The internal validations of LOO cross-validation,leave-many-out cross-validation,bootstrapping,and y-randomization test indicated that the QSTR model is robust and has no chance correlation. The external validations indicated that the model showed a good predict power. The QSAR model,therefore,can predict the unknown toxicity of organophosphorous pesticides.