Advanced chemometrics and fundamental considerations for non-targeted analysis with comprehensive multidimensional gas chromatography coupled with time-of-flight mass spectrometry

Abstract

Comprehensive two-dimensional gas chromatography (GC×GC) coupled with time-of-flight mass spectrometry (TOFMS) is a powerful analytical technique capable of separating complex mixtures, providing valuable information about the chemical composition of samples. However, the inherent data density associated with three-dimensional data provides a unique challenge to analytical chemists. As a result, significant effort has been invested in utilizing advanced chemometrics to glean meaningful information about samples from large and complex data sets. Herein, this dissertation introduces several investigations conducted on optimizing separation conditions to be amenable to chemometric deconvolution algorithms as well as the development, study, and application of advanced chemometric techniques applied to GC×GC-TOFMS data. To begin, the metric trilinear deviation ratio (TDR) is utilized to study the impact of experimental parameters such as column selection and modulation period, PM, on the quantitative accuracy of parallel factor analysis (PARAFAC) deconvolution. TDR scales with increasing change in second dimension retention time, Δ2tR, associated with pseudo-isothermal conditions on the second dimension, 2D, and quantitative accuracy decreases as TDR increases. Two column sets were utilized with varying film thickness on the first column, 1D, and each column set was studied using two PM for a total of 4 experiments. It was reported that using 1D columns with larger film thicknesses allows the analyst to employ a shorter PM, in turn lowering the Δ2tR, leading to higher quantitative accuracy. Many GC×GC-TOFMS studies relate to identifying class distinguishing analytes and can be tedious when performed manually. Fortunately, the use of discovery-based chemometric tools such as principal component analysis (PCA) and Fisher ratio (F-ratio) analysis has increased in popularity as less time-intensive and automated techniques for untargeted analyses. To begin, this dissertation will investigate mass channel purity obtained via the tile-based F-ratio algorithm using diesel fuel spiked with non-native analytes using GC×GC-TOFMS. The F-ratio algorithm, considered a supervised discovery technique because class membership is known a priori, was first used to “discover” the spiked non-native analytes. Then, using a novel signal ratio (S-ratio) algorithm, the mass channel selectivity information output by the F-ratio method was studied using three statistical metrics: null distribution analysis, p-value, and lack-of-fit (LOF). The result of this investigation revealed that a mass channel has a high likelihood of being pure when its p-value and LOF are sufficiently low. Finally, F-ratio analysis was applied to a dataset including patients with an anterior cruciate ligament (ACL) injury to discover potential biomarkers of post-traumatic osteoarthritis (PTOA) post-injury. Standard F-ratios are calculated by the between class variance divided by the sum of the within-class variance, scaling up as the between class variance increases and the within-class variance remains sufficiently small. However, many biological studies involve significant biological variance (~30%) that may not be associated with disease state or injury severity, etc. Herein, the standard tile-based F-ratio algorithm was modified to use only the within class variance associated with control samples. It was expected that the control class contained less within-class variance relative to the patient class, due to the expectation that some patient samples would be associated with increased severity of injury or the presence of coexisting conditions. Hit lists (metabolites discovered via F-ratio) from standard F-ratio and control normalized F-ratio were studied and directly compared to establish a comprehensive metabolome of potential biomarkers for PTOA development post ACL injury. Reported in this dissertation is a discussion on the complementary nature of standard and control-normalized F-ratio, followed by demonstration of class distinguishing metabolites via PCA.