Browsing by Author "Isomaa, Keijo"

This study focuses on chemometric analysis of instrumental data that has been obtained from chemical analysis of plant extracts. Chemometric analysis applies statistical and mathematical tools on chemical data, aiming to find new information or classifying samples in categories defined by the analyst. Chemometric analysis is based on computational pattern recognition and reveals any features that studied samples may have in common. In the literature part of this study, chemometrics and relevant concepts closely related to it are first explained and four commonly used chemometric methods are introduced, namely principal component analysis, hierarchical cluster analysis, k nearest neighbors and soft independent modeling of class analogy. The text is written with emphasis on being easily understandable without prior knowledge on the subject. After introducing these concepts, the literature concerning metabolomic studies of plant extracts published in the recent ten years are reviewed. This literature commonly employs chemometrics, aiming to discover if two or more varieties of the same plant species have markedly differing metabolomes and whether they can be exploited to automatically recognize these varieties. Additionally, the chemometric approaches often attempt to discover what factors are causing the successful findings. The purpose of the literature survey is to concretely show how chemometrics can achieve these goals, and to learn what the most common ways to treat the analytical data prior to chemometric analysis are. The experimental part applies chemometric methods to study bean extracts of the Ricinus communis plant, aiming to reveal if seed extracts of a same plant variety can be observed being similar, but clearly different from extracts of other varieties. Such situation could be exploited to develop a method that automatically identifies unknown seeds of the plant. The experimental work consisted of extracting homogenized samples with dilute aqueous acid, analyzing the extracts by three different instrumental techniques (liquid chromatography with ultraviolet light detection, liquid chromatography-mass spectrometry, and proton nuclear magnetic resonance spectroscopy) and finally analyzing the instrumental data by chemometric methods. Chemometrics research suffers from nonexistent standard operating procedures, since there is no universal way to treat a sample or data derived from it. While the main steps are often same, the details of sample preparation and preprocessing of analytical data vary greatly and can have a significant impact on the outcome. Despite, the data preprocessing is often left partially or completely manifested. The experimental finding was that six varieties of Ricinus communis could be successfully discriminated by both principal component analysis and hierarchical cluster analysis, applied on chromatographic data, while the results for spectroscopic data were not successful. The results encourage continuing the research, but with more emphasis on peak alignment and further experimenting with the preprocessing of the spectroscopic data. Choosing different short segments of the original spectroscopic profile is suggested, to leave out excessive information that is not helpful in discriminating the plant varieties but could obscure the relevant information.