【作者】 徐承建；

【导师】 梁逸曾；

【作者基本信息】 中南大学 ， 应用化学， 2003， 博士

【摘要】 现代分析仪器提供了大量而又丰富的量测信息，化学数据知识发掘的目的在于找到外在表征与内在结构之间的相互关系。高维数据解析方法的发展使人们对复杂化学体系分析能力有了很大的提高，使得传统分析化学难以处理的“黑色”体系的分析成为可能。本论文从当前化学学科发展现状出发，从色谱和质谱分析及其联用仪器入手，目的是通过建立能有效利用色谱和质谱提供的联合信息的化学计量学方法，从而实现对中药等复杂体系的定性和定量分析。论文主要涉及三个方面：质谱库数据发掘，联用色谱分辨方法和中药实际体系的分析。 一．质谱库数据发掘(第二章—第四章)：这部分主要是通过对质谱库的机器学习，来达到未知化合物的某些结构或子结构的预测。第二章提出了基于重采样技术的光谱特征选择方法，该法可找到包含某些子结构的特征信息，为质谱的分类和判别奠定了基础。第三章利用前人对质谱峰丰度统计的对数分布的先验知识，将原始数据进行合理的转化，降低了数据的方差，从而得到稳定的分类回归树。第四章采用调节样本权重的boosting技术来提高质谱数据的分类精度，有效降低了分类的误判率。 二．联用色谱数据分辨方法研究(第五章—第八章)：联用数据的分辨通常可分为两个问题：1)确定待测体系组分数 2)各个化学成分的光谱和色谱轮廓的求解。针对上述两个问题，第五章提出基于扰动的函数型主成分分析用于化学体系的秩估计，通过观察多次扰动采样来得到稳定的秩估计。第六章利用色谱的依次流出信息，提出了渐进窗口正交投影法来得到色谱和光谱轮廓，该法可有效避免局部共线性问题。针对特殊的包埋体系，第七章提出了诊断流出模式的渐进子窗口比较法。第八章提出了处理包埋体系的广义子窗口因子分析和基于最大嫡的估计方法。 三.实际中药分析应用研究(第九章一第十章):发掘中药化学背景，提高中医药的使用和疗效水平很大程度上依赖于对其活性化学组分协同作用的鉴别分析。本部分结合化学计量学分辨方法，主要研究了桂枝和鱼腥草两味中药的挥发油。第九章结合子窗口因子分析和渐进窗口正交投影法来实现对复杂体系的快速定性定量分析。第十章研究了四川产鱼腥草的挥发油，并将几个主要产地的鱼腥草挥发油主要成分进行了比较。更多还原

【Abstract】 The data mining in chemical data sets is to discover the hidden relationship between the chemical property and their structure. The recent development of high-dimensional chemometrics resolution methods provide a powerful tool for us to deal with the difficult "black system". Therefore, the aim of the thesis is to develop new methodology in mass spectra data mining and hyphenated chromatography research and apply the proposed methods to analyze the complex herbal medicines systems.1. Data-mining in mass spectra (chapter 2 to chapter 4). One can predict the presence or absence of the certain substructure by learning from mass spectra library. In chapter 2, a re-sampling technique is manipulated to obtain the characteristic spectra features. In chapter 3, an honest tree is built upon prior logarithm normal distribution. In chapter 4, a powerful boosting-tree is used to classify the mass spectra data sets with low prediction error.2. Hyphenated chromatographic resolution methods (chapter5 -chapter 8). In resolution problem, firstly, one should estimate the number of chemical components in the system under study, and then obtain the pure spectra and chromatographic profiles. In chapter 5, a noise perturbation method in functional principal component analysis is proposed to solve a chemical rank problem. In chapter 6, an evolving window orthogonal projection method is proposed to resolve the pure spectra and chromatographic profiles by utilizingthe evolving information in retention time direction. A new diagnosing method proposed in chapter 7 is to analyze the elution pattern of components. In chapter 8, a generalized sub-window factor analysis and entropy method is proposed to solve the embedded problem.3. Herbal medicine systems (chapter 9 to chapter 10): Building a new herbal medicine pharmacology theory usually heavily depends on the thorough understanding of their chemical knowledge background. The essential oil of Guizhi and Yuxingcao is well studied with chemometric resolution methods in this part. In chapter 9, a new evolving-approach is proposed to resolve complex Guizhi system. In chapter 10, we compare the main components of Yuxingcao form different region.更多还原