Chemical characterization of the acid alteration of diesel fuel: Non-targeted analysis by two-dimensional gas chromatography coupled with time-of-flight mass spectrometry with tile-based Fisher ratio and combinatorial threshold determination

Abstract

The illicit chemical alteration of petroleum fuels is of keen interest, particularly to regulatory agencies that set fuel specifications, or taxes/credits based on those specifications. One type of alteration is the reaction of diesel fuel with concentrated sulfuric acid. Such reactions are known to subtly alter the chemical composition of the fuel, particularly the aromatic species native to the fuel. Comprehensive two-dimensional gas chromatography coupled with time-of-flight mass spectrometry (GC×GC-TOFMS) is well suited for the analysis of diesel fuel, but may provide the analyst with an overwhelming amount of data, particularly in sample-class comparison experiments comprised of many samples. Tile-based Fisher-ratio (F-ratio) analysis reduces the abundance of data in a GC×GC-TOFMS experiment to only the peaks which significantly distinguish the unaltered and acid altered sample classes. Three samples of diesel fuel from differently branded filling stations were each altered to discover chemical features, i.e., analyte peaks, which were consistently changed by the acid reaction. Using different fuels prioritizes the discovery of features likely to be robust to the variation present between fuel samples and may consequently be useful in determining whether an unknown sample has been acid altered. The subsequent analysis confirmed that aromatic species are removed by the acid alteration, with the degree of removal consistent with predicted reactivity toward electrophilic aromatic sulfonation. Additionally, we observed that alkenes and alkynes were also removed from the fuel, and that sulfur dioxide or compounds that degrade to sulfur dioxide are generated by the acid alteration. In addition to applying the previously reported tile-based F-ratio method, this report also expands null distribution analysis to algorithmically determine an F-ratio threshold to confidently select only the features which are sufficiently class-distinguishing. When applied to the acid alteration of diesel fuel, the suggested per-hit F-ratio threshold was 12.4, which is predicted to maintain the false discovery rate (FDR) below 0.1%. Using this F-ratio threshold, 107 of the 3362 preliminary hits were deemed significantly changing due to the acid alteration, with the number of false positives estimated to be about 3. Validation of the F-ratio analysis was performed using an additional three fuels.

Performance evaluation of tile-based Fisher Ratio analysis using a benchmark yeast metabolome dataset

“…These data were completely identified and quantified previously and provide an insightful opportunity to compare the tile-based F-ratio analysis [29][30][31], which in turn had evolved from pixel-based F-ratio analysis [26][27][28]. The novelty of the study herein is based upon the simultaneous demonstration of improved F-ratio analysis software performance for a complex metabolomics dataset, which is manifested as a decisive improvement in the ranking of true positive hits above false positive hits, in concert with the demonstration that the null distribution threshold accurately identifies this optimized transition from true positive hits to false positive hits in the hit list.…”

“…While GC × GC-TOFMS is an outstanding instrumental platform for biological and metabolomics studies [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26], there is an ongoing need to develop software methods to glean useful information from the immensely complex data. For this purpose Fisher Ratio (F-ratio) analysis has been found to be a particularly useful algorithm for the analysis of these complex datasets [27][28][29][30][31].…”

“…Points with high F-ratios generally correspond to features which distinguish between the classes compared. Previously, F-ratio analysis using GC × GC data was applied to several biological and petrochemical models [26][27][28][29][30][31]. The prior pixel-based F-ratio studies were fruitful, and provide a benchmark for current software development and evolution.…”

“…To address this challenge of determining an appropriate threshold to prudently guide hit list data mining, combinatorial null distribution analysis is coupled with tile-based F-ratio analysis [30,31], leveraging the large volume of related data that may be tested by GC × GC-TOFMS to experimentally determine the distribution of potential false positives. By rearranging the sample classes to nullify the class distinguishing variance, the effects of non-meaningful variation in the dataset may be estimated for the true class comparison.…”

“…There are 200 possible different null arrangements of a six versus six comparison and this number rapidly increases as the sample class population increases. Each replicate data subset was subjected to F-ratio analysis of both the one true class comparison for metabolite discovery and the additional 200 null classes for threshold determination [30,31]. Four replicate data subsets were analyzed in order to assess whether or not results from any given individual subset adequately represent the whole dataset of thirty-five R versus thirty-five DR chromatograms [25,26].…”