摘要
在DFT-B3LYP/6-311+G(d,p)水平对60种非环状亚硝胺分子结构进行几何全优化,通过多元逐步线性回归(MSR)分析筛选出9个量子化学描述符作为自变量,log LD_(50)(lethal dose 50%,LD_(50):大鼠口服急性毒性)作为因变量,采用人工神经网络(ANN)方法构建QSAR模型。经Levenberg-Marquardt(LM)算法训练得到的隐含层为10个神经元节点的多层感知机ANN模型为最优结构。采用内外双重验证的方法,分析和检验模型的稳健性。对模型的内部验证采用留一法(LOO)交叉验证和均方根误差(RMSE)评估,其结果为Q_(LOO)^(2)=0.9514,RMSE_(train)=0.1534;外部验证结果为Q_(ext)^(2)=0.8842,RMSE_(test)=0.2948,因此构建的QSAR模型具有优秀的预测能力,为非环状亚硝胺结构和其急性毒性关系的深入理解提供理论基础。
In this study,a QSAR model was developed by the LD50 values of 60 acyclic NAs on rats using Artificial Neural Network(ANN)to predict the acute oral toxicity of other NAs.The geometric structures of all acyclic NAs were optimized using Gaussian 09 program package DFT-B3LYP/6-311+G(d,p)basis set and totally 34 quantum chemistry descriptors were obtained,such as molecular dipole,atomic charges,energetic data and so on.Using a multiple stepwise linear regression(MSR)method,9 descriptors which were important to the LD_(50) of acyclic NAs on rats,were chosen to establish the QSAR model.The optimal results were obtained with 10 hidden layer neuron nodes ANN model trained with the Levenberg-Marquardt(LM)algorithm.The accuracy of the predicted model for the internal training set was evaluated by leave-one-out(LOO)cross-validation(CV)and root mean square error(RMSE)with Q_(LOO)^(2) by 0.9514 and RMSE_(train) by 0.1534.For further external test,Q_(ext)^(2) and RMSE_(test) are 0.8842 and 0.2948,respectively.Overall,the model developed in this study can be used to predict the acute oral toxicity of acyclic NAs and provide some useful information for better understanding the relationship between the chemical structural and the acute oral toxicity of NAs.
作者
陈雅菲
钟儒刚
白洁
CHEN Ya-fei;ZHONG Ru-gang;BAI Jie(The Green Aerotechnics Research Institute of Chongqing Jiaotong University,Chongqing 401120,China;Faculty of Environment and Life,Beijing University of Technology,Beijing 100124,China)
出处
《化学研究与应用》
CAS
CSCD
北大核心
2022年第1期58-67,共10页
Chemical Research and Application
基金
重庆交通大学绿色航空技术研究院自设项目(GATRI2020D02003)资助。
