Advances in the Chemometric Analysis of Multiway Chromatographic Data to Improve Discovery and Identification

Abstract

Both one-dimensional gas chromatography (1D-GC) and comprehensive two-dimensional gas chromatography (GC×GC) are used in various applications because of their ability to discover and identify pure chemical species in volatile and semi-volatile mixtures. However, the information-rich data sets produced by these instruments, especially when they are coupled to mass spectrometry (MS), are often too large and complex to manually interpret. Therefore, the application of advanced data analysis methods, referred to as chemometrics, are necessary to efficiently analyze and extract meaningful chemical information from these instrumental platforms. This dissertation presents the development and application of several novel chemometric approaches that improve the discovery and identification of key chemical species in both 1D-GC-MS and GC×GC-MS data sets. First, this dissertation describes the application of Fisher ratio (F-ratio) analysis to create a chemical fingerprint of potato taste defect in roasted coffee beans and thermal stress in kerosene-based rocket fuels. As a supervised chemometric technique, F-ratio analysis utilizes prior knowledge of sample class membership to discover statistically significant concentration differences in chromatographic data sets. However, knowledge about the samples or experimental design may not be available during analysis. To address this situation, this dissertation describes the development of two unsupervised data analysis approaches. For large chromatographic data sets, variance ranking analysis was created to discover analytes exhibiting a high signal variance across the samples. Application of variance ranking analysis, along with principal components analysis and k-means clustering, to multiple metabolomic data sets uncovered hidden chemical patterns and sample groupings. Variance ranking analysis was also demonstrated to be an effective data reduction technique for developing accurate physicochemical models of aerospace fuels with partial least squares regression. For studies that may be limited in the number of samples and/or chromatographic replicates, a pairwise analysis method known as 1v1 analysis was developed to find chemical differences between two chromatograms. This method can also extract a purified mass spectrum to improve compound identifications for analytes at low chromatographic resolutions and/or with high signal interferences. The performance of both unsupervised analyses was shown to be comparable to F-ratio analysis. Finally, this dissertation also advances the capacity to reliably discover and identify analytes using a single chromatogram. The generation of an enhanced total ion current chromatogram (TIC) is introduced to improve visualization of analytical signals previously obscured by the background noise. The enhanced TIC algorithm improves the detection of analytical signals by denoising the mass spectral dimension. Concurrently, an intra mass channel (m/z) comparison method, termed mzCompare, is developed to improve the identification of unresolved chemical species. This approach generates pure analyte profiles for unresolved chemical species by discovering m/z with similar retention times and peak shapes. These purified profiles are then used as a constraint in a chemometric decomposition model to mathematically resolve the overlapped species and achieve accurate compound identifications.