Research And Analysis Of The Relationship Between The Structure And The Property Of Substituted Benzenes And Liquid Crystal And Styreue Polymerization

QSPR/QSAR(quantitative structure-property relationships , quantitative strueture-activity relationships) have been a cutting-edge research in chemistry and environmental science.QSPR/QSAR(quantitative structure-property/activity relationships)quantitatively describes the relationships between the molecular structures of compounds and their activities/physicochemical properties by means of the mathematical models on the basis of structural parameters of compounds. Making use of the mathematical models, we can predict the activities and physicochemical properties of some unknow compounds.In our dissertation, we researched the basal theoretics of molecular structure's quantitative description and quantitative structure-property/activity relationships and summarized the application on QSAR/QSPR method in chemistry and the environmental science. In addition, we mainly discussed a novel machine learning method, radial basis function neural networks (RBFNN) to construct QSPR/QSAR models and made some evaluation on the established models.In chapter one, the dissertation included a brief description of the history and fundamental and method and applications of QSAR/QSPR. In this section, we introduced some regression methods (HM, RBFNN and SVM) and made a review on their application.In chapter two, the dissertation described detailedly the application of HM and RBFNN and SVM in chemistry and environmental chemistry. A brief introduction was given as follows:1. Our paper used QSTR to deal with modeling of the acute toxicity of 92 substituted benzenes. Heuristic method (HM) and radial basis function neural networks (RBFNNs) were utilized to construct the linear and the nonlinear QSTR models, respectively. The predictive results were in agreement with the experimental values. The optimal QSTR model which was established based on RBFNNs gave a correlation coefficient (R~2) of 0.893, 0.876, 0.889 and root-mean-square (RMS) error of 0.220, 0.205, 0.218 for the training set, the test set and the whole set, respectively. RBFNNs proved to be a very good method to assess acute aquatic toxicity of these compounds. The current model allows a more transparent chemical interpretation of the acute toxicity in terms of intermolecular interactions. Furthermore, the proposed approach can also be extended in other QSAR investigations.2. Quantitative structure-property relationships (QSPRs) models have beensuccessfully developed for the prediction of the nematic transition temperatures (T_N)of 42 thermotropic liquid crystals. Heuristic method (HM) and radial basis function neural networks (RBFNNs) and support vector machine (SVM) were utilized to construct the linear and nonlinear QSPRs models, respectively. Furthermore, by comparing the results from different QSPR approaches, it can be seen that the nonlinear RBFNNs model can describe more accurately the relationship between the structural parameters and the property. The optimal QSPRs model which was established based on RBFNNs gave a square correlation coefficient (R~2) of 0.984, 0.953, 0.973 and root-mean-square (RMS) error of 2.19, 4.13, 2.99 for the training set, the test set and the whole set, respectively. Some analysis to the data set and evaluation were done in the paper. It can also provide an idea for dealing with QSPRs problem of thermotropic liquid crystals.3. Quantitative Structure-Reactivity Relationships (QSRRs) were used to study kinetic chain-transfer constants for 90 transfer agents on styrene polymerization at60℃. Heuristic Method (HM) and Radial Basis Function Neural Networks (RBFNNs)method were utilized to construct the linear and nonlinear prediction models, respectively. Comparing the results obtained from the two models, the prediction results of RBFNNs model were much better than that obtained by HM. The RBFNNs model gave a square correlation coefficient R~2 of 0.8842 for the training set, 0.8586 for the test set, 0.8908 for the whole data set, respectively. The HM provides a transparent result for modeling the reactivity which indicates that hydrogen bond and charge factor are closely related to the reactivity. Additionally, the training procedure is also simple when using nonlinear RBFNN model based on the same set of descriptors. The detailed validation procedure that contained in the models (separation of the data into two independent sets, cross-validation) illustrated the accuracy and robustness of the produced models not only by calculating its fitness on the training data, but also by testing its predicting ability. Furthermore, by comparison the results from different QSRR approaches, it can be seen that the nonlinear RBFNN model can explain the factors affecting reactivity from mechanistic hypotheses.