Quantitative Structure-activity Relationship Studies Of The Toxicological Properties Of Benzene Compounds

There are about10million different chemicals known in the world, andmeanwhile, thousands of new chemicals are developed each year. Most of thechemicals are harmful to human health and environment. Therefore, it is an importanttopic in the field of environment protection to judge the toxicological hazards of theseorganic chemicals scientifically and objectively. Currently, with the development ofcomputer technology and statistical methods, the quantitative structure-activityrelationship (QSAR) studies has been used widely in modern chemistry, biochemistry,and so on. Its research object includes biological activities, toxicity, pharmacokineticparameters, bioavailabilities of compounds, environmental behavior and a variety ofphysical and chemical properties of the molecular. QSAR models can reveal therelationship between structure of the compounds and its activity from the molecularlevel, people can use it to design, screen and predict the toxicity characters of theinterested compounds.Benzene compounds are the important raw materials used extensively in thechemical, pharmaceutical and other industries. They also contribute to the chemicalsthat caused potential environmental and health problems. Therefore, we took theBenzene compounds as the research object, combining the molecular structure of thequantitative description with multiple linear regression (MLR) or non-linearregression statistical methods support vector machine (SVM), to build successfullythe acute toxicity QSAR models and mutagenic QSAR models of Benzenecompounds. This work will provide the technical support for the environmental andhealth risk assessment of new chemicals.Chapter1of the dissertation included a brief description of the QSAR history,basic principles, methods and research progress, and a simple introduction of the basicprinciple of several kinds of statistical methods and its application in QSAR models.In chapter2, we used the quantitative structure and activity relationship methodsto build the multiple linear QSAR models of acute toxicity for eight classes of organiccompounds (benzoic acid, benzoic acid esters, aldehydes, ketones, nitrobenzene,amino benzene, phenol, alkyl benzene, halogenated benzene). The results of modelanalysis suggested that the models have good prediction ability and stability on theacute toxicity of the benzene compounds. In chapter3, we used the quantitative structure and activity relationship methodsto build the multiple linear QSAR models of mutagenicites for three classes oforganic compounds (nitrobenzene, amino benzene, halogenated benzene class). Theresults of model analysis suggested that the models have good prediction ability andstability on the mutagenicites of the benzene compounds.In chapter4, we introduced a new machine learning method named supportvector machine (SVM) in QSAR modeling. We selected the acute toxicity data ofamino benzene class to construct nonlinear QSAR model successfully with theapplication of SVM algorithm. By comparison with the linear QSAR models, wefound that nonlinear QSAR models created by SVM algorithm has a better stabilityand prediction ability than that of multiple linear QSAR model.In conclusion, the QSAR models for the selected organic compounds based onMLR and SVM were built in this dissertation. The built models can be used to predictthe toxicity property of these compounds, and reveal the action mechanism ofenvironmental pollutants. In addition, by comparing the linear and nonlinear QSARmodels, we suggest that the stability and prediction ability of nonlinear QSAR modelsare better than that of multiple linear QSAR models.