The Applications Of Artificial Neural Networks In The Separation System Of Chromatography And Capillary Electrophoresis

Capillary electrophoresis (CE) and chromatography along with distillation, recrystallization, solvent extraction, chemical deposition, electrolytic deposition and centrifugal separation are the important separation and analytical techniques in Chemistry. After the long-term development, Capillary electrophoresis and chromatography separation techniques have been two of the most important branches of Analytical Chemistry. With the development of modern science and technology, newly devised multi-dimensional hyphenated chromatography and capillary electrophoresis instruments have been applied widely in Chemistry and Analytical Chemistry. So it is possible and necessary to obtain a great amount of measured chemical data. Since 1970's, obtaining a great amount of data has not been the most difficult step in chemical analysis and measurements. Moreover, more attention should be paid to the extraction of the useful information from the large amount of measured chemical data and data analysis. Meanwhile, the same problem arose in the separation and analysis field by capillary electrophoresis and chromatography. To resolve the multi-dimensional data measured from complex system, computer-aided complex mathematical methods for data analysis are necessary. However, data and signal processing method originated from other subjects have also been introduced to the analysis for the complex data measured in chemical process. Hence, Chemometrics was founded in 1970's.Artificial neural networks is an important branch of Chemometrics. For itspowerful projection ability for nonlinear models, artificial neural networks can be applied in nonlinear process modeling, multivariate calibration and resolution in nonlinear system and pattern recognition analysis in Chemistry. In the separation system of capillary electrophoresis or chromatography, generally speaking, nonlinear process (or system) is more universal than linear one. Moreover, sometimes, it is difficult to found an explicitly defined mathematical model that is suitable for modeling some process (or system) in capillary electrophoresis or chromatography. Under the conditions, it is suitable and necessary to introduce -artificial neural networks to the corresponding study on the separation system of capillary electrophoresis and chromatography. At present, comprehensive investigation and study on the topic have been reported.Nevertheless, it is necessary to point out that although the great progress has been made in the study of artificial neural networks, the investigation for its theory and applications are both still focused on the preliminary stage. Many problems of artificial neural networks still need to be studied further, such as the training rate of networks, overfitting, the structure and the modeling ability of networks, etc. All these problems include some uncertainties. Hence, in this research work, artificial neural network was applied in the separation system of capillary electrophoresis or chromatography. The main research area of this dissertation includes modeling of the retention or migration behavior in CE or chromatography, quantification of the corresponding components in overlapped CE peaks and clinical pattern recognition analysis for data set generated fromcapillary electrophoresis or chromatography analysis process. In the corresponding research work, different input selection strategies were employed to improve the modeling ability of corresponding artificial neural networks and simplify the structures of the artificial neural networks. Meanwhile, the modeling abilities of different kinds of artificial neural networks in some separation systems were compared.Three main sections were included in this research work: 1. Modeling of the retention or migration behavior in chromatography orcapillary electrophoresis 1). Modeling of the relationship between electroosmotic flow and correspondingseparation parameters in capillary zone electrophoresisThe prediction of migration time of electroosmotic flow (EOF) marker was achieved by applying artificial neural networks (ANN) model based on principal component analysis (PCA) and standard normal distribution simulation to the input variables. The voltage of performance, the temperature in the capillary, the pH and the ionic strength of background electrolytes (BGE) were applied as the input variables to ANN. The range of the performance voltage studied was from 15kV to 27kV, and that of the temperature in the capillary was from 20 ℃ to 30 ℃. For the pH values studied, the range was from 5.15 to 8.04. The range of the ionic strength investigated in this paper was from 0.040 to 0.097. The prediction abilities of ANN with different pre-processing procedure to the input variables were compared. Under the same performance conditions, the average predictionerror of the migration time of the EOF marker was 5.46% with RSDn-1=1.76% according to ten parallel runs of the optimized ANN structures by the proposed approach, and that of the ten parallel predictions of the optimal ANN structures for the different performance conditions was 12.95% with RSDn-1=2.29% according to the proposed approach. The study showed that the proposed method could give better predicted results than other approaches discussed.2). The study of the relationship between the new topological index Am and the gas chromatographic retention indices of hydrocarbons by artificial neural networksThe newly developed topological indices Am1-Am3 and the molecular connectivity indices mX were applied in structure-property correlation studies. The topological indices calculated from the chemical structures of some hydrocarbons were used to represent the molecular structures. The prediction of the retention indices of the hydrocarbons on three different kinds of stationary phase in gas chromatography can be achieved applying artificial neural networks and multiple linear regression models. The results from the artificial neural networks approach were compared with those of multiple linear regression models. It is shown that the predictive ability of artificial neural networks is superior to that of multiple linear regression method under the experimental conditions in this paper. Both the topological indices 2X and Am1 can improve the predicted results of the retention indices of the hydrocarbons on the stationary phase studied.2. Quantification for the corresponding components in their overlapped capillary electrophoresis peaks1). Quantification in overlapped capillary electrophoresis peaks based on different kinds of artificial neural networksThe application of four different kinds of artificial neural networks (ANN) for quantification of overlapped peaks in micellar electrokinetic capillary chromatography (MECC) is investigated. The four kinds of ANN applied in this section were: linear ANN model, radial basis function ANN model, generalized regression ANN model and multiple layer perceptron ANN model. In the case of overlapped CE peaks, the four kinds of ANN were proved to be promising approaches for quantification of the corresponding components. Both the spectra and the electrophoretograms of the unseparated analytes were used as the multivariate input data. The two kinds of the data were both suitable for quantification of overlapped peaks by the four different ANN models. In the study, it was also shown that the linear ANN models could give the best predicted results for the corresponding components in their partially overlapped MECC peaks.2). Artificial neural networks based on genetic input selection for quantification in overlapped capillary electrophoresis peaksThe application of multilayer perceptron artificial neural networks (MLP ANN) based on genetic input selection for quantification of the unresolved peaks in micellar electrokinetic capillary chromatography (MECC) is reported. An optimization strategy for genetic input selection was also proposed. When thecorresponding CE peaks cannot (or are difficult to) be resolved completely only by separation techniques, MLP ANN based on genetic input selection can be a suitable tool to resolve the problem. Both the spectra and the electrophoretograms of the unseparated analytes were used as the multivariate input data. The two kinds of the data were both suitable for quantification of overlapped CE peaks by MLP ANN based on genetic input selection. The study also shows that the applying of genetic input selection in MLP ANN can improve the precision of quantification in both completely and partially overlapped CE peaks to some extent. 3). Artificial neural networks based on principal component analysis input selection for quantification in overlapped capillary electrophoresis peaksThe application of three different kinds of artificial neural networks (ANN) (radial basis function ANN, generalized regression ANN and linear ANN) based on principal component analysis (PCA) input selection for quantification of overlapped peaks in micellar electrokinetic capillary chromatography (MECC) is investigated. In the case of overlapped CE peaks, ANN based on PCA input selection were proved to be a promising approach for quantification of the corresponding components. Both the spectra and the electrophoretograms of the unseparated analytes were used as the multivariate input data to the corresponding ANN based on PCA input selection. The two kinds of the data were both suitable for quantification of overlapped peaks by ANN based on PCA input selection. In the study, it was also shown that the input selection procedure based on PCA for the three kinds of ANN could improve the precision of quantification of thecorresponding components in both completely and partially overlapped CE peaks to some extent.3. Pattern recognition analysis in the separation system of CE or chromatography 1). Artificial neural networks based on principal component analysis input selection for clinical pattern recognition analysisTwo clinical data sets were applied for pattern recognition analysis in order to discover the correlation between urinary nucleoside profiles and tumors. One data set contains 24 clinical urinary samples, of which 12 specimens are from female thyroid cancer patients (malignant tumor group), and the other samples belong to healthy women (normal group). However, 28 clinical urinary samples are comprised in the second data set. In all the specimens, each number of the samples for both uterine cervical cancer patients (malignant tumour group) and healthy females (normal group) are 10, and the other 8 samples are from uterine myoma patients (benign tumour group). For the two data sets, the clinical urinary nucleoside profiles analyses were performed by capillary electrophoresis (CE) method. The pattern recognition analysis was achieved applying multiple layer perceptron artificial neural networks (MLP ANN) based on conjugate gradient descent training algorithm. Moreover, the precision of the results of the pattern recognition analysis was improved to some extent (or without any deterioration) even by much simpler structure of MLP ANN when the strategy of input selection based on principal component analysis (PCA) to MLP ANN was applied. The study showed that MLP ANN based on PCA input selection was a promising tool forpattern recognition analysis.2). Application of artificial neural networks in clinical pattern recognition analysis: a comparative study for different input selection strategiesMultiple layer perceptron artificial neural networks (MLP ANN) based on conjugate gradient descent training algorithm was employed to perform the pattern recognition analysis for a clinical data set generated from gas chromatography analysis. 26 clinical samples were collected, of which 12 were from uterine myoma patients, and the others belong to uterine cervical cancer patients. In each sample, 50 kinds of clinical urinary organic acids were determined. Applying the input selection strategies based on PCA, stepwise regression analysis, forward regression analysis, backward regression analysis and genetic algorithm, the pattern recognition analysis results of the corresponding ANN models were improved to some extent, and the structures of the ANN models were simplified. The study showed that for the clinical data set investigated in this research work, ANN models based on stepwise or forward regression analysis input selection strategies can give more satisfied pattern recognition analysis results.