The Construction Of In Silico Chemical Mutagenicity Models And Quantitative Regressive Workflow

Computational toxicology has become an essential assisted tool in the research of chemical toxicology.This thesis contained the introduction of the current development of computational toxicology, and how I applied the computational toxicology methodologies to predict the mutagenicity of the compound. Moreover, how to use KNIME to develop the workflow of the QSAR was also included. As a result, we usedworkflow to develop models and applied these models to predict the Tetrahymena pyriformis toxicity and aqueous solubility.The first chapter of the thesis described the background and the application of Computational toxicology. The content included the expert systems, statistic models, web server, machine learning methods, and fingerprints.The chapter2mainly showed the research work of building the in silico predictive models of chemical mutagenicity, which is one of the most important endpoints of toxicity. The research comprised several steps as follows. First, collecting the mutagenicity data of compounds, and then calculating the chemical fingerprints. Afterward, constructing the models by using the machine learning methods. The last stage was to evaluate the performances of the models. Finally, we got three best models(PubChem-kNN,MACCS-kNN and PubChem-SVM) whose predictive accuracies were all above90%.In the third chapter of the thesis, we introduced the methodologies of using KNIME to build the workflow of quantitative structure-activity relationship. As is known to all, KNIME is free of charge and widely used as a tool of workflow. In our work, we firstly employed KNIME to invoke PaDEL-Descriptor software to calculate the MACCS fingerprint. After that, we made use of the support vector regression to build the quantitative models. Then, we applied this methodology to predict the Tetrahymena pyriformis toxicity and aqueous solubility of compounds. The correlation coefficient p2was larger than0.5, which proved that the quantitative models had the ability to predict properties.Chapter4was the summary of the whole thesis, and the prospect of the computational toxicology.