化合物毒性预测模型构建及烟草烟气化学成分毒副作用预测研究

发布时间：2018-07-22 14:45

【摘要】：化合物毒性是导致药物研发失败的主要原因之一。将药物安全性评价提前到药物研发早期阶段有助于缩短药物研发周期,降低开发成本。除了药物之外,食品添加剂、化妆品等与生活相关的其他化学品的安全性评估也是十分必要的。常规的毒理学试验方法需要耗费大量的时间和金钱,难以满足现代药物研发以及环境化合物风险评估的要求。研究者迫切需要依据“3R”原则开发替代方法,即减少(Reduce)实验动物数量、优化(Refine)实验程序、替代(Replace)实验动物。使用计算机方法进行化合物毒性预测,可以用较低的成本实现对大批量化合物进行快速的安全性评价。计算机方法的另一个优势是对药物发现早期阶段的虚拟化合物,也可以进行毒性预测。因此,计算毒理学方法已经成为化合物安全性评价中不可或缺的工具。本论文的研究工作主要包括两大部分：第一部分是化合物毒性预测模型构建,在这一部分工作中,分别构建了化合物急性毒性和致癌毒性的预测模型,第二部分是烟草与烟气化学成分毒副作用研究,在这里构建了烟草与烟气中的化学成分数据库,并对其进行了毒副作用预测研究,另外,使用系统评价方法分析了吸烟对于帕金森症患病风险的影响。论文的第一章,首先介绍了化合物毒性预测研究的相关背景和意义,以及目前计算机预测化合物毒性的主要方法和技术,包括交叉参照、定量／定性构-毒关系、警示子结构等。然后,介绍了国内外有关烟草与烟气及其有害成分的研究进展。接下来,介绍了帕金森症以及吸烟对帕金森症发病率的研究背景和系统评价研究进展状况。另外,也对用于系统评价的(?)neta分析(Meta-analysis)方法进行了介绍。化合物的急性毒性是在药物研发以及生态风险评估过程中必须要考虑的一个重要毒性性质。动物实验方法测定化合物的急性毒性需要大量的时间和资金投入,计算机预测是一种很重要的替代方法。论文第二章中,我们收集整理了12204种化合物的结构及其大鼠口服急性毒性实验数据。按照US EPA的毒性分级标准,将所有化合物划分为剧毒、高毒、中毒和微毒四个等级。然后,使用MACCS与FP4两种分子指纹表征分子结构,结合支持向量机、κ-最近邻居法、C4.5决策树、随机森林和朴素贝叶斯等五种机器学习方法分别构建了多分类模型。其中,支持向量机本身是一个两类问题的判别方法,我们采用“一对一”与“二叉树”两种策略用于多分类问题的解决。采用两个分别包含1678和375个化合物的外部验证集对构建的多分类模型进行了验证。模型MACCS-SVMOAo表现出了最强的预测能力,对两个外部验证集的总体预测准确率分别达到83.0%和89.9%。另外,我们采用信息增益技术和子结构频率分析方法,筛选了可能导致化合物产生急性毒性的优势子结构片段。致癌性是另一种受到广泛关注的化合物毒性性质。论文第三章,我们提取了CPDB数据库中829种化合物的大鼠致癌性实验数据,使用五种机器学习方法结合六种分子指纹类型,建立了用于预测化合物致癌毒性的二分类和三分类预测模型。使用测试集对模型进行初步验证后,分别选取了8个表现最好的二分类模型和7个表现最好的三分类模型,使用外部验证集进行进一步验证。外部验证集由ISSCAN数据库中的87种化合物构成。预测能力最强的二分类模型MACCS-kNN整体预测准确率达到83.91%,预测能力最强的三分类模型MACCS-kNN整体预测准确率也达到80.46%。我们筛选了五种在致癌化合物中出现频率明显高于非致癌物的结构片段,为化合物的致癌性研究提供警示作用。烟草烟气是严重危害人体健康的的成分十分复杂的混合物质。对烟草与烟气中的化学成分进行汇总和分析,有助于香烟风险评估研究以及监管香烟生产过程中有害成分的检测。论文第四章,我们构建了一个烟草与烟气化学成分数据库(The Chemical Components of Tobacco and Tobacco Smoke, CCTTS)。 CCTTS提供了烟草与烟气中5983种化学成分的化学结构、基本物理化学性质、毒性信息以及使用admetSAR工具预测得到的ADMET(吸收,分配,代谢,排泄和毒性)性质。在这些化学成分中,568种已有明确的实验毒性数据,另有145种化学成分被预测为具有急性毒性或者致癌性。本数据库将可通过互联网进行浏览和检索。吸烟有害人体健康,不过许多流行病学研究表明吸烟有可能会降低患帕金森症的风险。论文的第五章,我们通过meta分析方法系统评价了现有的吸烟与帕金森症风险的研究结果。我们全面地检索了1960年到2014年10月期间有关吸烟与帕金森症风险相关性研究的英文文献,从中筛选了61项回顾性的病例对照研究和9项前瞻性的队列研究。与不吸烟者相比,吸烟者患帕金森症的相对风险值RR(95%置信区间)的合并值为0.59(95%置信区间,0.56-0.62),表示曾经吸烟者患帕金森症的风险比不吸烟者低41%。我们进一步按照不同的流行病学研究方法、受试者性别、对照组来源、吸烟量以及研究年代等进行了亚组分析。各个亚组的meta分析,都表现出了吸烟在帕金森症方面的保护作用,这种保护性作用十分明显,并且吸烟量越大,这种保护作用越强。我们总结了目前有关这种保护作用生物机制的一些假说,建议研究者们深入了解这一保护作用的生物机制,进而有助于从烟草和烟气成分中发现潜在的延缓和治疗帕金森症的药物。论文的第六章对全文的研究工作进行了总结,并突出了论文的创新点。
[Abstract]:The toxicity of compounds is one of the main causes of the failure of drug research and development. The early stage of drug safety assessment to the early stage of drug development can help to shorten the cycle of drug development and reduce the cost of development. Besides, it is also necessary to evaluate the safety of other chemicals, such as food additives, cosmetics and other related chemicals. Regulatory toxicology test methods need to spend a lot of time and money. It is difficult to meet the requirements of modern drug development and environmental compound risk assessment. Researchers urgently need to develop alternative methods based on the "3R" principle, that is, to reduce the number of experimental animals (Reduce), to optimize (Refine) experimental procedures, to replace (Replace) experimental animals. The computational method can be used to predict the toxicity of compounds, which can be used to evaluate the safety of large quantities of compounds at a lower cost. The other advantage of the computer method is to detect the virtual compounds at the early stage of the drug discovery and to predict the toxicity. Therefore, the method of computational toxicology has become an evaluation of the safety of compounds. The research work of this paper consists of two major parts: the first part is the construction of the compound toxicity prediction model. In this part, the prediction model of acute toxicity and carcinogenicity of compounds is constructed, the second part is the study of the side effects of tobacco and flue gas, and the tobacco is constructed here. The chemical composition database in the flue gas is used to predict the toxic side effects and the effects of smoking on the risk of Parkinson's disease. In the first chapter, the background and significance of the study of compound toxicity prediction are introduced, and the computer prediction of the toxicity of compounds at present is introduced. The main methods and techniques, including cross reference, quantitative / qualitative toxic relationship, warning sub structure, etc., then introduced the research progress on tobacco and smoke and its harmful components at home and abroad. Then, the research background and systematic evaluation of the incidence of Parkinson's disease and smoking on the incidence of Parkinson's disease are introduced. The (?) NETA analysis (Meta-analysis) method used for system evaluation is introduced. The acute toxicity of compounds is an important toxic nature to be considered in the process of drug development and ecological risk assessment. Animal experimental methods require a large amount of time and capital to determine the acute toxicity of compounds. Computer prediction is a kind of method. In the second chapter, we collected and collated the structure of 12204 compounds and their rat oral acute toxicity test data. According to the toxicity grading standard of US EPA, all the compounds were classified into four levels of toxic, high toxicity, poisoning and microtoxicity. Then, two molecular fingerprints of MACCS and FP4 were used to characterize molecular structures, Combined with support vector machines, kappa nearest neighbor method, C4.5 decision tree, random forest and naive Bayes, the multi classification model is constructed respectively. Among them, support vector machine itself is a discriminant method of two kinds of problems. We use the "one to one" and "two forked tree" two strategies to solve the multi classification problem. Using two external validation sets containing 1678 and 375 compounds, the multi classification model is validated. The model MACCS-SVMOAo shows the strongest prediction ability, and the overall prediction accuracy for two external validation sets is 83% and 89.9%. respectively. We use the information gain technology and the substructure frequency analysis method, The predominant substructure fragment which may lead to the acute toxicity of the compound is screened. Carcinogenicity is another kind of toxic properties which are widely concerned. In the third chapter, we extracted the experimental data of the rat carcinogenicity of 829 compounds in the CPDB database, combined with five kinds of machine learning methods to combine six types of molecular fingerprints. Two classification and three classification prediction models used to predict the carcinogenicity of compounds. After using the test set to verify the model, 8 two classification models with the best performance and 7 best three classification models are selected, and the external validation sets are used for further verification. The external validation set is composed of 87 kinds of combination in the ISSCAN database. The two classification model with the strongest prediction ability, the overall prediction accuracy of MACCS-kNN, is 83.91%, and the overall prediction accuracy of the three classification model, MACCS-kNN, which is the most powerful, has also reached 80.46%.. We screened five types of carcinogenic compounds in the carcinogenic compounds. For warning. Tobacco smoke is a complex compound that seriously endangering human health. The summary and analysis of chemical components in tobacco and smoke contribute to the study of cigarette risk assessment and the detection of harmful components in the process of cigarette production. In the fourth chapter, we build a tobacco and flue gas chemistry. The component database (The Chemical Components of Tobacco and Tobacco Smoke, CCTTS). CCTTS provides the chemical structure of 5983 chemical components in tobacco and flue gas, basic physicochemical properties, toxicity information, and ADMET (absorption, distribution, metabolism, excretion and toxicity) predicted by the use of admetSAR tools. In these chemical components, 5 68 kinds of clear experimental toxicity data, and another 145 chemical components are predicted to be acute or carcinogenic. This database will be accessible and retrieved through the Internet. Smoking is harmful to human health, but many epidemiological studies suggest that smoking may reduce the risk of Parkinson's disease. The fifth chapter of the paper, we The current study of smoking and Parkinson's risk was systematically evaluated by the meta analysis. We retrieved the English literature on the risk of smoking and Parkinson's disease during the period from 1960 to October 2014, and screened 61 retrospective case control studies and 9 prospective cohort studies. Compared to smokers, the relative risk value of Parkinson's disease RR (95% confidence interval) was 0.59 (95% confidence interval, 0.56-0.62), indicating that smokers had a risk of suffering from Parkinson's disease than non smokers and 41%. we further followed different epidemiological methods, subjects sex, control sources, smoking, and age of study. Meta analysis in each subgroup showed the protective effect of smoking on Parkinson's disease. This protective effect was obvious, and the greater the amount of smoking, the stronger the protective effect. We summarized some hypotheses about the biological mechanism of this protective effect, and suggest that researchers understand this one in depth. The biological mechanism of protection helps to discover potential drugs for delay and treatment of Parkinson's disease from tobacco and smoke components. The sixth chapter of the paper summarizes the research work of the full text, and highlights the innovation of the paper.
【学位授予单位】：华东理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R99

【相似文献】