苯酚及其衍生物对水生梨形四膜虫急性毒性的QSTR研究

发布时间：2018-03-09 14:47

本文选题：苯酚及其衍生物　切入点：定量构效关系　出处：《山西师范大学》2015年硕士论文　论文类型：学位论文

【摘要】：目前，随着合成工业的不断发展，苯酚及其衍生物被广泛地用作化工原料。因其具有明显的生物毒性、内分泌干扰性和生物富集作用，成为普遍存在的污染物，对人类及其它动植物带来一定的危害，这就使对其生态毒性进行风险评估的研究有了一定的价值。随着研究的深入，为了节省时间和金钱，了解毒性机理，模拟非合成化合物的生态毒理学行为，同时满足公众反对动物测试的呼吁，定量结构-毒性相关（Quantitative Structure-Toxicity Relationship，QSTR）研究在环境科学领域已取得了多项成果，，故有必要开展以QSTR方法建立苯酚及其衍生物毒性模型的研究。在20世纪90年代后期，发现了一种测试生物毒性的新方法——梨形四膜虫毒性测试法。该测试方法检测速度快、方法操作简单、经济实惠、应用的范围广范，因而在药物、有机物、无机物、水污染物等方面广泛用于毒理学评价。本文在查阅大量文献的基础上，选取了258个苯酚及其衍生物对水生梨形四膜虫的毒性数据，结合软件筛选出的7个分子描述符作为建模的结构参数，分别采用多元线性回归（MultipleLinear Regression，MLR）、偏最小二乘（Partial Least Squares，PLS）、BP（BackforwardPropagation，BP）神经网络三种方法进行定量构效关系研究。本文主要工作包括：（1）介绍了本文运用的相关化学计量学方法，如多元线性回归、偏最小二乘、BP神经网络、主成分分析（Principal Component Analysis，PCA）等的基本原理。（2）利用ADMEWORKS ModelBuilder软件（Version4.5Standard）计算并筛选了258个苯酚及其衍生物的分子描述符，最后选出7个描述符作为建模变量。（3）运用稳健诊断方法（Robust Diagnostic Method）剔除24个奇异样本，继而采用球排除算法（Sphere-exclusion Algorithms）将样本合理划分为多个训练集与内部测试集，同时要求内部测试集和外部验证集均匀分布在PC空间的整个区域内，最终划分为3类合理的样本。（4）结合取自文献的毒性数据，分别采用多元线性回归、偏最小二乘、BP神经网络方法进行QSTR研究，成功建立了毒性预测模型，并对外部验证集采用共识建模方法（Consensus Modeling Method），从而提高模型的预测能力。根据QSTR研究结果分析表明，所建模型均具有较好的预测能力和稳定性，且与MLR、PLS模型相比，BP神经网络模型性能略胜一筹，即非线性模型比线性模型性能优越。但是BP神经网络建立的模型不能直接给出直观的数学模型和公式，而MLR、PLS模型更为简单明了。本文建立的QSTR预测模型，实现了只需知道苯酚及其衍生物的分子结构式，而不用做实验就可以合理预测其毒性数据。
[Abstract]:At present, with the development of synthetic industry, phenol and its derivatives are widely used as chemical raw materials. It brings some harm to human beings and other animals and plants, which makes the study of ecological toxicity risk assessment have some value. As the research goes on, in order to save time and money, to understand the mechanism of toxicity, Mimicking the ecotoxicological behavior of non-synthetic compounds and satisfying the public's appeal against animal testing, quantitative structure-toxicity related Structure-Toxicity relationships (QSTRR) studies have made many achievements in the field of environmental science. Therefore, it is necessary to establish the toxicity model of phenol and its derivatives by QSTR method. In late 1990s, a new method for toxicity testing of Tetrahymena piriformis was discovered. The method is fast, easy to operate, economical, and widely used in drugs. Organics, inorganic substances and water pollutants are widely used in toxicological evaluation. Based on a large number of literatures, 258 toxic data of phenol and its derivatives to Tetrahymenum piriformis were selected. Using seven molecular descriptors selected by the software as the structural parameters of modeling, the quantitative structure-activity relationships were studied using three methods of multivariate linear regression and partial Least squarestio (partial Least squarestio) neural network, respectively, and BP BackforwardPropagation (BP) neural networks. The main work of this paper includes:. This paper introduces the basic principles of the related chemometrics methods, such as multivariate linear regression, partial least squares BP neural network, principal component analysis (PCA) and so on. The molecular descriptors of 258 phenol and its derivatives were calculated and screened by ADMEWORKS ModelBuilder software version 4.5 Standard, and 7 descriptors were selected as modeling variables. The robust Diagnostic method is used to eliminate 24 strange samples, and then the spherical exclusion algorithm is used to divide the samples into multiple training sets and internal test sets. At the same time, it is required that the internal test set and the external verification set are uniformly distributed in the whole area of the PC space, and finally divided into three categories of reasonable samples. In combination with the toxicity data obtained from literature, the QSTR model was successfully established by using multivariate linear regression and partial least squares BP neural network method. Consensus Modeling method is applied to the external validation set to improve the prediction ability of the model. According to the analysis of QSTR research results, the proposed model has good predictive ability and stability, and the performance of BP neural network model is better than that of MLRN model. That is, the performance of nonlinear model is superior to that of linear model, but the model established by BP neural network can not give direct mathematical model and formula, while MLR QSTR model is more simple and clear. Only the molecular structure of phenol and its derivatives is known, and the toxicity data can be reasonably predicted without experiment.
【学位授予单位】：山西师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ243.12;TQ086.51

【参考文献】