| Gas chromatography mass spectrometry (GC-MS), which combines the high-performance separation capability of gas chromatography and abundant structural information from mass spectrometer, has been widely used to analyze the volatile or semi-volatile compounds. So far, in the processing of the GC-MS data, the main methods of qualitative analysis are similarity search in the reference mass spectral library and/or comparison of the retention index with standards or ones in library. Since there are many highly similar mass spectra in the reference mass spectral library, it is very difficult to determine the correct result from hit list. In this case, the accuracy of identification greatly reduces for most of analysts simply take the first candidate with the highest similarity as the correct result. More importantly, the method of library searching does not apparently well work for the unknown compounds out of library. To address these questions, three strategies are proposed as follows:1) mine the reference mass spectra and retention indices to obtain the mass spectral characteristics and elution features, and then combine these two information to determine the structure for unknown compounds; 2) establish the special library of mass spectra and retention indices for in-house analysis; 3) develop calibration model of mass spectra from different sources and distinguish isomers with highly similar mass spectra and highly similar retention indices. In quantitative analysis, with the help of discreteness of mass spectra, we propose two methods to identify the potential selective ions, whose chromatogram could be used to initialize concentration vector for iterative target transformation factor analysis. The main contents of this thesis are:1. With the help of data mining and stepwise refinement, the mass spectral classifier is built by the neutral mass loss and characteristic ions. Additionally, dissimilarity analysis is developed and employed to build the prediction machine of number of branches in 2-heptene and 2-octene. This method calculates the difference spectrum of the unknown spectra and the corresponding spectra of normal 2-heptene or 2-octene, and then selects the most abundant peak in the difference spectrum as important ion, which is used to calculate the neutral loss. Finally, the neutral loss and some characteristic ions are emplyed to build the prediction machine. Through comparison with the chemometric methods of pattern recognition, the results indicate that the classifier and structure prediction machine built by neutral loss and characteristic ions have the following advantages:a) the better classification or prediction results can be achieved; b) the model can easily be expanded to similar compounds, due to important variables have been explained by the classical theory of fragmentation; c) neutral loss is ingeniously described and employed in mass spectrometry classification and feature extraction for the first time. (Chapter II)2. After the mass spectral characteristics are obtained with the help of dissimilarity analysis, they are used to combine the retention indices of monomethylalkane to develop new method of qualitative analysis. In this thesis, firstly, mass spectral characteristics of alkane are summarized using the data mining and stepwise refinement and employed to establish model to identify the normal alkane and other alkanes; secondly, dissimilarity analysis of standard mass spectrometry of 90 monomethyl-alkanes is carried out to discover the local features of mass spectra, which could be employed to deduce the methyl position. Combining these features with retention index and the molecular ion peak, an automated method of identifying monomethylalkane is developed. In this method, complementary left and right concave, retention index and molecular ion peak could work together to improve the accuracy of the qualitative results; thirdly, through data mining of reference mass spectra of wax ester, the several mass spectral characteristics are summarized and explained by classic theory of fragmentation. According to the profiling of the whole spectra, the wax could be classified into three categories in the light of unsaturation degree in two moieties. For these three kinds of mass spectra, the calculation formulas are designed by mass of characteristic ions. Tested by the simulated data, the real data and validated by standards, the results indicate that the dissimilarity analysis is an effective method to mine the mass spectral characteristics and the combination of mass spectral characteristics and retention index could improve the accuracy of qualitative analysis for unknown compounds. (Chapter III and IV)3. The strategy of special database is proposed and demonstrated by a simple library of fatty acids. To avoid the chromatographic shift, the fixed experimental conditions are used to analyze all samples. The chemometrics methods are employed to solve the overlapping peaks to overcome the shortage of fixed conditions. After the establishment of retention indices and mass spectra database, a method combining the mass spectral characteristics, retention indices and chemometrics is proposed to analyze the fatty acids in biological samples without standards. Furthermore, five edible vegetable oils, Eucommia ulmoides seed oil and fatty acids in human plasma samples are analyzed by this method. The saturated fatty acids are firstly identified by mass spectral characteristics and then used to calculate the equivalent chain length (ECL) values for other wax esters. Finally, the comparison with the ECL values in the special data is carried out to identify the unsaturated fatty acids. The results show that Eucommia ulmoides seed oil has nutritional value and this method could provide the fatty acid profiling of human serum for the diagnosis of disease with a solid basis for qualitative and quantitative analysis. (Chapter V)4. Two calibration methods to transfer the mass spectra detected in different instruments or under different conditions to target are developed and employed to distinguish isomers with highly similar mass spectra and highly similarly retention indices. Two methods are the vector of calibration ratios and piecewise direct standardization. The first one mainly computes the weighted average of the ratio of relative abundance, while the second one is the piecewise direct standard with calibration method of least squares regression. After tested by a training set and two types of test sets, two methods are proved to be effective for not only for mass spectra in the training set, but also for fatty acids with different branch, functional groups, double bond position and the geometrical isomerisation of double bonds. For cis/trans isomers of polyunsaturated fatty acid methyl esters, the differences among their mass spectra are smaller than systematic difference from instrument or condition. After calibration, the most of the geometrical isomerisation of double bonds could be identified. (Chapter VI)5. Two detectors of potential selective ion are proposed and used to initialize the concentration vector of iterative target transformation factor. The former method compares the start and end points determined from evolving factor analysis (EFA) and mass chromatograms, while the latter one analyzes the difference of mass spectra from different regions to identify potential selective ion. For multi-component system segmentation strategy is proposed to divide the whole data into several sub-windows, in which just one component mainly exists. The concentration profiles could be obtained from each subwindow and then be used to calculated mass spectra with the help of the least squares regression, whose standard spectra could be employed to compute the pure chromatographic curves by using the least squares regression. Through compared with previous method and tested by the low SNR data system, volatile oil of Eclipta prostrata,97# gasoline, the results suggest that the two methods could quickly get higher quality analytical results. (Chapter VII)... |
PDF Full Text Request