Tile-Based Pairwise Analysis of GC × GC-TOFMS Data to Facilitate Analyte Discovery and Mass Spectrum Purification

Abstract

A new tile-based pairwise analysis workflow, termed 1v1 analysis, is presented to discover and identify analytes that differentiate two chromatograms collected using comprehensive two-dimensional (2D) gas chromatography coupled with time-of-flight mass spectrometry (GC × GC-TOFMS). Tile-based 1v1 analysis easily discovered all 18 non-native analytes spiked in diesel fuel within the top 30 hits, outperforming standard pairwise chromatographic analyses. However, eight spiked analytes could not be identified with multivariate curve resolution-alternating least-squares (MCR-ALS) nor parallel factor analysis (PARAFAC) due to background contamination. Analyte identification was achieved with class comparison enabled-mass spectrum purification (CCE-MSP), which obtains a pure analyte spectrum by normalizing the spectra to an interferent mass channel (m/z) identified from 1v1 analysis and subtracting the two spectra. This report also details the development of CCE-MSP assisted MCR-ALS, which removes the identified interferent m/z from the data prior to decomposition. In total, 17 out of 18 spiked analytes had a match value (MV) > 800 with both versions of CCE-MSP. For example, MCR-ALS and PARAFAC were unable to decompose the pure spectrum of methyl decanoate (MVs < 200) due to its low 2D chromatographic resolution (∼0.34) and high interferent-to-analyte signal ratio (∼30:1). By leveraging information gained from 1v1 analysis, CCE-MSP and CCE-MSP assisted MCR-ALS obtained a pure spectrum with an average MV of 908 and 964, respectively. Furthermore, tile-based 1v1 analysis was applied to track moisture damage in cacao beans, where 86 analytes with at least a 2-fold concentration change were discovered between the unmolded and molded samples. This 1v1 analysis workflow is beneficial for studies where multiple replicates are either unavailable or undesirable to save analysis time.

Recent Advances in GC×GC and Chemometrics to Address Emerging Challenges in Nontargeted Analysis

“…141 However, Cain et al recently demonstrated that the principles of CCE-MSP can also be modified for situations where there is only a single chromatogram per class. 115 In this application, pure interference m/z are identified using the LOF along with the RM (vide supra) in place of p-values. Even with this replacement in the identification of pure interference m/z, CCE-MSP still demonstrated superior performance, especially for analytes with low R s,2D and high s int /s A , relative to MCR-ALS and PARAFAC.…”

“…Figure C illustrates the development of a tile-based approach for supervised, pairwise comparisons for GC×GC-TOFMS data, termed 1v1 analysis . Tile-based 1v1 analysis uses the same four-grid tiling scheme originally developed for tile-based F -ratio analysis (see Figure ), and recently applied for unsupervised analysis with VRI-USI (see Figure D).…”

“…This method was originally demonstrated by Ochoa et al using the experimental design and outputs from tile-based FRA . However, Cain et al recently demonstrated that the principles of CCE-MSP can also be modified for situations where there is only a single chromatogram per class . In this application, pure interference m / z are identified using the LOF along with the RM (vide supra) in place of p -values.…”

“…Figure 8C illustrates the development of a tile-based approach for supervised, pairwise comparisons for GC×GC-TOFMS data, termed 1v1 analysis. 115 Tile-based 1v1 analysis uses the same four-grid tiling scheme originally developed for tile-based F-ratio analysis (see Figure 4), and recently applied for unsupervised analysis with VRI-USI (see Figure 5D). However, 1v1 analysis calculates a Rank Metric (RM), defined as the sum-normalized absolute difference between the two chromatograms, extending comparisons to sample-limited and/or run-limited studies.…”

Tile-Based Random Forest Analysis for Analyte Discovery in Balanced and Unbalanced GC × GC-TOFMS Data Sets

“…In this respect, multiple F-ratio amended calculations were suggested, such as the control-normalized F-ratio ( CN FRA) and the minimum variance optimized F-ratio ( MVO FRA). Furthermore, multiple studies investigated combining the tiling approach with other metrics, such as relative variance (RSD 2 ) in the tile-based variance rank initiated-unsupervised sample indexing (VRI-USI) method and the use of a rank metric (RM) for pairwise analysis workflows . Despite the extensive efforts devoted to optimizing and extending the tile-based FRA to accommodate various settings, the scenario of unbalanced data sets, i.e., different numbers of samples per class, remains untreated and calls for further attention.…”