This report is about applying a Fisher ratio method to entire four dimensional (4D) data sets from third-order instrumentation data. The Fisher ratio method uses a novel indexing scheme to discover the unknown chemical differences among known classes of complex samples. This is the first report of a Fisher ratio analysis procedure applied to entire 4D data sets of third-order separation data, which, in this case, is comprehensive two-dimensional gas chromatography coupled with time-of-flight mass spectrometry analyses of metabolite extracts using all of the collected mass channels. Current analysis methods for third-order separation data use only user-defined subsets of the 4D data set. First, in a validation study, the Fisher ratio method was demonstrated to objectively evaluate and determine the chemical differences between three controlled urine samples that differed by known spiked chemical components. It was determined that, out of more than 600 recognizable chemical components in a single sample, the six spiked components, along with only two other matrix components, differed most significantly in concentration among the control samples. In a second study, the Fisher ratio method was used in a novel application to discover the unknown chemical differences between urine metabolite samples from pregnant women and nonpregnant women. A brief list of the top 11 components that were most significantly different in concentration between the pregnant and nonpregnant samples was generated. Because the Fisher ratio calculation statistically differentiates regions of the chromatogram with large class-to-class variations from regions containing large within-class variations, the Fisher ratio method should generally be robust against biological diversity in a sample population. Indeed, application of principal component analysis in this second study failed due to biological diversity of the samples.

A comprehensive two-dimensional (2D) retention time alignment algorithm was developed using a novel indexing scheme. The algorithm is termed comprehensive because it functions to correct the entire chromatogram in both dimensions and it preserves the separation information in both dimensions. Although the algorithm is demonstrated by correcting comprehensive two-dimensional gas chromatography (GC x GC) data, the algorithm is designed to correct shifting in all forms of 2D separations, such as LC x LC, LC x CE, CE x CE, and LC x GC. This 2D alignment algorithm was applied to three different data sets composed of replicate GC x GC separations of (1) three 22-component control mixtures, (2) three gasoline samples, and (3) three diesel samples. The three data sets were collected using slightly different temperature or pressure programs to engender significant retention time shifting in the raw data and then demonstrate subsequent corrections of that shifting upon comprehensive 2D alignment of the data sets. Thirty 12-min GC x GC separations from three 22-component control mixtures were used to evaluate the 2D alignment performance (10 runs/mixture). The average standard deviation of first column retention time improved 5-fold from 0.020 min (before alignment) to 0.004 min (after alignment). Concurrently, the average standard deviation of second column retention time improved 4-fold from 3.5 ms (before alignment) to 0.8 ms (after alignment). Alignment of the 30 control mixture chromatograms took 20 min. The quantitative integrity of the GC x GC data following 2D alignment was also investigated. The mean integrated signal was determined for all components in the three 22-component mixtures for all 30 replicates. The average percent difference in the integrated signal for each component before and after alignment was 2.6%. Singular value decomposition (SVD) was applied to the 22-component control mixture data before and after alignment to show the restoration of trilinearity to the data, since trilinearity benefits chemometric analysis. By applying comprehensive 2D retention time alignment to all three data sets (control mixtures, gasoline samples, and diesel samples), classification by principal component analysis (PCA) substantially improved, resulting in 100% accurate scores clustering.