摘要
运用HF/6-31G~*、HF/6-311G^(**)、DFT-B3LYP/6-31G~*、DFT-B3LYP/6-311G^(**)方法全优化计算18种酚类化合物,获得相应的量子化学参数:最高占有轨道能(E_(HOMO))、最低空轨道能(E_(LUMO))、前沿轨道能级差(△E=E_(LUMO)-E_(HOMO))、分子偶极距(μ)、各原子净电荷(q_i)、苯环上净电荷增量△Q_R和分子体积(V)等。利用多元线性回归分析(MLR)和人工神经网络误差反传算法(BP)2种数据挖掘方法,研究酚类化合物对黑曲霉抑制毒性的定量构效关系(QSAR),采用去一法(LOO)交互检验的方法验证模型稳健性和预测能力,选出最佳模型。所建最佳MLR和BP模型的相关系数、LOO交互检验复相关系数分别为0.956、0.834和0.991、0.845,所建QSAR模型的稳健性和预测能力良好。结果表明酚类化合物对黑曲霉的抑制毒性与分子体积、最低空轨道能、苯环上净电荷增量的相关性较好;分子体积越大,化合物毒性越大;E_(LUMO)越低,毒性越大;△Q_R增大,苯环上的正电性增强,亲电性愈强,化合物毒性愈大。
The HF/6-31G^＊, HF/6-311G^＊＊, DFT-B3LYP/6-31G^＊3DFT-B3LYP/6-311G^＊＊ were employed to calculate the molecular geometric and electronic strucres of 18 phenols. EHOMO, ELLO, △E, μ, qi,△QR and V Were selected＇as Structural descriptors. The inhibiton toxicity of such phenols to aspergillus niger along with the above structural parameters was used to establish the quantitative structure-activity relationship （QSAR） models by multiple linear stepwise regressions （MLR）. The estimation stability and generalization ability of these models were analyzed by leave-one-out method and the best one was selected. Meanwhile, standard back-propagation algorithm of artificial neural network （BP） was used to establish a nonlinear QSAR model. The correlation coefficient （R） of established MLR and BP models, leave-one-out （LOO） eross validation Rcv, are 0.956, 0.834 and 0.991, 0.845, respectively. These show that QSAR models have both favorable estimation stability and good prediction capability. The results indicate that the inhibiton toxicitie was increased with the increase of the volume of the compound and the increment of the net charge of the benzene ring, and decreased with the increase of the ELUMO.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2011年第1期107-110,共4页
Computers and Applied Chemistry
基金
江苏省滩涂生物资源与环境保护重点建设实验室开放基金资助课题(JLCBE07019)
盐城师范学院自然科学研究项目(09YCKL008)
关键词
酚类化合物
黑曲霉
定量构效关系
多元线性回归
人工神经网络误差反传算法
phenols, aspergillus niger, quantitative structure-activity relationship, multiple linear stepwise regression, standard back-propagationalgorithm of artificial neural network
