【作者】 王岚；

【导师】 卢小泉； 康敬万；

【作者基本信息】 西北师范大学 ， 分析化学， 2007， 硕士

【副题名】基于几种模式识别方法在目标分子与DNA相互作用中的应用

【摘要】 金属类配合物和抗生类化合物(简称目标分子)作为抗癌药物的研究是一个有意义而有待深入研究的课题。它们与DNA相互作用的研究一直是人们关注的焦点;开展这方面的研究对于DNA分子和其选择性作用机理的进一步认识、活性药物的体外筛选、分子设计、新药的合成和治癌药物的治癌机理等均有重要的意义。分子与DNA的作用模式、作用机理、结合位点、键合常数是此类研究的主要内容。目前,荧光光谱、紫外.可见吸收光谱、质谱、毛细管电泳和电化学等方法都被用来分析DNA和目标分子的相互作用,并取得了一定的进展。但仪器方法在这类研究中存在两个问题,一是不能批量预测目标分子与DNA作用的常数和模式,一是不能从分子的结构角度分析相互作用的影响因素。本论文以主成分分析(principal components analysis(PCA))、人工神经网络(artificial nerves net(ANN))、多元统计方法(multivariate analysis(MLR))和系统聚类分析法(hierarchical cluster analysis(HCA))等为研究工具,对目标分子与DNA的相互作用与影响因素进行了研究。本文研究工作分为以下几个部分:第一部分绪论讲述了模式识别方法(主成分分析、人工神经网络和系统聚类法)的发展简史,介绍了这些分析方法的基本数学原理,并分别综述了它们在分析化学中的应用。第二部分金属类配合物与DNA相互作用的预测性研究金属配合物和DNA相互作用的研究是一项具有很高的理论价值和实用价值的工作;研究金属配合物和DNA相互作用的方式和机理,总结其相互作用的一般规律,对探索配合物在抗肿瘤药物、分子生物学、生物工程技术及其它相关领域的应用具有非常重要的意义。本文选择了63种金属配合物作为研究对象,从分子水平讨论了金属配合物与DNA相互作用的主要影响因素。应用主成分分析方法讨论了显著影响相互作用的因素;同时用多元线性回归和人工神经网络建立了两个预测作用常数的模型和一个预测作用模式的模型。初次量化的分子结构参数有21种,经过筛选发现其中的10种参数对相互作用有显著的影响。所建立的模型对作用模式和常数的预测有较好的准确性。第三部分抗生类化合物与DNA相互作用的化学计量学研究抗生类化合物与DNA相互作用的研究在了解药物作用机理、分子设计和药物筛选等方面有重要意义。目前主要以实验方法研究两者的作用机理(包括作用常数和作用模式),较难从分子结构水平发现影响相互作用的因素,而且对于批量的未知化合物与DNA的相互作用无法进行预测。本实验以抗生类化合物与DNA相互作用的实验数据和目标分子的结构量化数据为基础,应用主成分分析、系统聚类分析、多元线性回归和人工神经网络方法讨论了显著影响相互作用的因素,建立了两个预测作用常数的模型和一个预测作用模式的模型。初次量化的分子结构参数有24种,经过筛选发现其中的12种对相互作用有显著的影响。对于键合常数的预测,最小相对误差只有0.17%(MLR)和0.72%(ANN);而对于键合模式的预测,五个测试分子与实验结果完全一致。第四部分模型对于杂环类分子与DNA相互作用常数的预测根据前面工作中得出的预测作用常数的模型,本部分实验通过该模型预测了30种杂环类目标分子与DNA相互作用的作用常数。同时,通过已发现的显著影响因素讨论了这些参数对杂环类目标分子与DNA相互作用的影响。这些理论研究为药物分子的设计和筛选可提供有价值的信息。更多还原

【Abstract】 The study of metal compounds and antibiotics (target molecules) as anti-cancer drug was a meaning and remarkable topics. The interaction of these target molecules and DNA was always a foucs that people paied close attention to. An understanding of how molecules bind to DNA would be potentially useful in the design and filtering new drugs, medicine synthesizes therapy and genetic diseases, etc.The binding mode, function mechanism, binding site point and binding constant are the main contents of these kinds of study. Up to now, fluorescence spectrum, ultraviolet-visible spectroscopy, capillary electrophoresis and electrochemical techniques have been developed for the identification and characterization of interactions between them, and have made great progress. However, there are still some problems in this field. First, it is difficult to find the influencing factors from the structure of molecules. Second, for batches of unknown molecules, it is difficult to predict the interaction from the experimental results.In this paper, principal components analysis, multivariate analysis, artificial neural network and hierarchical cluster analysis were applied to study the interaction of target molecules and DNA and their influencing factors.The main works of this paper-were as follows:Part I: IntroductionThis part related the developing history of chemical pattern recognition techniques (principal components analysis, artificial nerves net and hierarchical cluster analysis), introduced their basic mathematical theory and summed up their application in the field of analytical chemistry. Part II: Predictive studies on interaction between DNA and metal compoundsThe study of the interaction of metal compounds and DNA has very high theories worth and practical value. The way and the mechanism of the interaction of DNA and metal compound, and their general regulation were very useful to investigate the application of compound in anti-tumor medicine, molecular biology, technique and other related field of the living creature engineering.In this paper, 63 metal compounds were selected as research object to discuss main influencing factors of their interaction with DNA from molecular structure. Principal components analysis (PCA) was employed in order to reduce dimensionality and investigate which subset of variables could be more effective for classifying the compound according to their binding mode with DNA. Artificial nerves net (ANN) and multivariate analysis (MLR) were applied to establish two predicting constant models and one predicting binding mode model. 21 parameters (descriptors) that can influence the interaction of selected metal compound and DNA have been investigated. Results of the PCA study showed that there are 10 descriptors closely related to the interaction. The proposed models had the good accuracy in predicting constant and mode.Part III: Studies on the interaction between antibiotics and DNAInteraction between antibiotic molecules and DNA has the important significance on understanding drug function, designing molecular and filtering drugs. The present studies aimed to find interaction mechanism by experimental techniques. But it is difficult to find the structure influence factors that affect the interaction, and for batches of target molecular, the interaction information is difficult to predict.In this study, based on experimental data and quantified structure data, principal components analysis, ANN, HCA and MLR were applied to study the interaction. The influence factors were investigated. Two models of predicting interaction constant and one model of predicting interaction mode were established. Among 24 quantified parameters, it was found that there were 12 parameters that affected the interaction markedly. For the prediction of binding constant, the minimal relative error is only 0.17% (MLR) and 0.72% (ANN); for the prediction of binding mode (ANN), five test molecules are all consistent with experimental results.Part IV: Predict the interactive constant of DNA and heterocyclic molecules by modelAccording to the predictive constant model by previous studies, 30 binding constant of the interaction of heterocyclic molecules (dye compounds) and DNA were predicted in this study. Meanwhile, we discussed the effect of the interaction of target molecule by the influenced factors which had been found. The investigations in this work provide worthy information to design and filter medicine molecules.更多还原