Chemometric techniques for modeling and classification of composition and identity in multivariate analytical chemical data

Quantitative and qualitative analyses are traditional analytical tasks that are more effectively conducted using multivariate statistical methods. Modern spectroscopic, chromatographic, and mass spectrometric methods produce data that are difficult to interpret, because of both their complexity and high dimensionality. Multivariate techniques, including principal component analysis (PCA), classification and regression trees (CART), hierarchical cluster analysis (HCA), and canonical variate analysis (CVA) are used to reduce the dimensionality of such large data sets and to discover rules for discriminating different sample types.; PCA is used extensively as a preliminary step for data visualization and dimensionality reduction. In an experiment designed for undergraduate instrumental analysis, mixtures of xylene isomers were discriminated from one another by PCA of their UV spectra; the composition of a mixture of xylene isomers was estimated by principal component regression (PCR).; GC/MS analyses were performed on mixtures of gasoline, kerosene and diesel accelerants according to a simplex mixture design. PCR to estimate composition of accelerant mixtures from the GC/MS profiles was based on principal components that best correlated with accelerant composition; excellent calibrations were achieved for all three accelerants.; CART is a nonparametric method that uses recursive binary partitioning to select the best splitting rules for classification of samples into pre-assigned groups. CART identified combinations of ion masses that differentiate mass spectra of alditol hexaacetates from one another. Pyrolysis chromatograms from two types of mouse fibroblasts (normal and virally transformed) were also analyzed by CART; a simple rule based on a single chromatographic peak was found to differentiate the two cell lines. Chromatograms obtained by sampling the headspace of brewed coffee poured in two different containers were classified using a combination of multivariate techniques. Although the sets of chromatograms were difficult to distinguish reliably by visual inspection, PCA, CART, and HCA were able to differentiate the two groups.; Mass spectra of xylene isomers obtained using both quadrupole and a time-of-flight mass spectrometry were analyzed by CART, PCA, and CVA. While PCA could be used to visualize groupings, classification of these isomers based solely on their mass spectra was possible with CART and CVA.