Tile-based Fisher ratio analysis of comprehensive two-dimensional gas chromatography time-of-flight mass spectrometry (GC × GC-TOFMS) data using a null distribution approach.

Comprehensive two-dimensional (2D) gas chromatography coupled with time-of-flight mass spectrometry (GC × GC-TOFMS) is a versatile instrumental platform capable of collecting highly informative, yet highly complex, chemical data for a variety of samples. Fisher-ratio (F-ratio) analysis applied to the supervised comparison of sample classes algorithmically reduces complex GC × GC-TOFMS data sets to find class distinguishing chemical features. F-ratio analysis, using a tile-based algorithm, significantly reduces the adverse effects of chromatographic misalignment and spurious covariance of the detected signal, enhancing the discovery of true positives while simultaneously reducing the likelihood of detecting false positives. Herein, we report a study using tile-based F-ratio analysis whereby four non-native analytes were spiked into diesel fuel at several concentrations ranging from 0 to 100 ppm. Spike level comparisons were performed in two regimes: comparing the spiked samples to the nonspiked fuel matrix and to each other at relative concentration factors of two. Redundant hits were algorithmically removed by refocusing the tiled results onto the original high resolution pixel level data. To objectively limit the tile-based F-ratio results to only features which are statistically likely to be true positives, we developed a combinatorial technique using null class comparisons, called null distribution analysis, by which we determined a statistically defensible F-ratio cutoff for the analysis of the hit list. After applying null distribution analysis, spiked analytes were reliably discovered at ∼1 to ∼10 ppm (∼5 to ∼50 pg using a 200:1 split), depending upon the degree of mass spectral selectivity and 2D chromatographic resolution, with minimal occurrence of false positives. To place the relevance of this work among other methods in this field, results are compared to those for pixel and peak table-based approaches.