Application Of Radial Basis Function Neural Networks In Environmental Chemistry And Medical Chemistry

Quantitative structure-activity/property Relationship (QSAR/QSPR) methods are the most promising and successful tools to provide rapid and useful meaning for predicting the biological activity or toxicity of organic compounds by using of different statistical methods and various kinds of molecular descriptors. The aim of QSAR is to develop models on a training set of compounds, these models will then allow for the prediction of the biological activity of related chemicals. This kind of study can not only develop a method for the prediction of the property of compounds that have not been synthesized but also can identify and describe important structural features of molecules that are relevant to variations in molecular properties, thus gain some insight into structural factors affecting molecular properties. Now, QSAR method has been introduced to environment chemistry and medical chemistry. In this dissertation, we mainly discussed a novel machine learning method, radial basis function neural networks (RBFNN) to construct QSAR model.Chapter 1 of the dissertation included a brief description of the history, principle, realization process and research status of QSAR. In this section, we also introduced the method RBFNN and a review of the application of RBFNN in medical and environment chemistry area.Chapter 2 of this dissertation described the application of RBFNN in environment chemistry. In resent years, lots of chemical were used in industrial and agricultural areas and the by-products were discharged into the environment. Nearly all the compounds have potential toxic effect to human being and the whole biosysterm. QSAR models can be used to predict the toxicity of most organic compounds for the risk assessment. Meanwhile, QSAR methods saved time and cost. A brief introduction was given as follows:(1) Quantitative structure-retention relationship (QSRR) models have been successfully developed for the prediction of the retention factor (log k) in the biopartitioning micellar chromatography (BMC) of 66 organic pollutants. Heuristic method (HM) and RBFNN were utilized to construct the linear and non-linear QSRR models, respectively. The RBFNN model gave a correlation coefficient (R~2) of 0.8464 and root-mean-square error (RMS) of 0.1925 for the test set. This work provided a useful model for the predicting the log k of other organic compounds when experiment data are unknown.(2) This paper presents the results of an optimization study on the toxicity of 91 aliphatic and aromatic compounds as well as a small subset of pesticides to algae Chlorella vulgaris, which was accomplished by using QSAR. The linear method (HM) and the nonlinear method (RBFNN) were used to develop the QSAR models and both of them can give satisfactory prediction results: the root-mean-square errors (RMS) were 0.4023 and 0.3124, respectively. At the same time, by interpreting the descriptors, we can get some insight into structural features related to the toxic action. Additionally, the diversity analysis of the compounds was completed in the selected features space and a detailed analysis on the model application domain defined the compounds, whose estimation can be accepted with confidence. The results of this study suggest that the proposed approaches could be successfully used as a general tool for the estimate of novel toxic compounds. It is very valuable for the risk assessment in environment science.Chapter 3 described the application of QSAR and RBFNN method to medical chemistry. The main were generalized as follows:(1) HM and RBFNN methods were proposed to generate QSAR models for a set of non-benzodiazepines ligands at the benzodiazepines receptor (BzR). Descriptors calculated from the molecular structures alone were used to represent the characteristics of the compounds. Compared with the results of HM, the RBFNN obtained more accurate prediction. The correlation coefficients (R) of the nonlinear RBFNN model were 0.9113 and 0.9030 for the training and testing sets, respectively. This paper proposed an effective method to design new ligands of BzR based on QSAR.(2) The inhibitory activity to epidermal growth factor receptor (EGFR) tyrosine kinase for 61 analogues of 4-(3-bromoanilino)-6,7-dimethoxyquinazoline were modeled with the descriptors calculated from the molecular structure alone using a QSAR technique. The multiple linear regression (MLR) and RBFNN were utilized to construct the linear and nonlinear prediction models. The prediction result of the RBFNN model is more accurately than that of MLR. For the test set, the predictive correlation coefficient R of 0.8688 and 0.9030 were obtained. It has been demonstrated that RBFNN is a powerful tool for QSAR prediction. However, the MLR linear model could give some insight into the factors that are likely to govern the inhibitory activity of the compounds and be used as an aid to the drug design process.