Simulating GC×GC-TOFMS data & the robustness of chemometric software (Timothy Trinklein, MDCW 2023)

• 🎤 Presenter: Timothy Trinklein, Robert Synovec (University of Washington, Seattle, USA)

💡 Book in your calendar: 15th Multidimensional Chromatography Workshop (MDCW) January 2024

15th Multidimensional Chromatography (MDC) Workshop 2024

Abstract

We report a workflow to simulate GC×GC-TOFMS datasets with realistic run-to-run retention time shifting which are used to evaluate the robustness of chemometric software to shifting.

By using simulated data, chromatographic modeling variables such as the amount of shifting, crowdedness (i.e., saturation), and peak area RSD% could be systematically varied. Data were shifted between runs by using low frequency “shift functions”, which were compared to the shifting observed in a yeast metabolomic study to ensure the results were generalizable to real data. We focused on the evaluation of tile-based F-ratio analysis (FRA), a supervised, non-targeted analysis tool which segments chromatograms into “tiles” and uses the ANOVA F-ratio to identify regions containing class-distinguishing analytes, output into a table (i.e., "hit list").

A broad range of simulation parameters were tested to thoroughly evaluate the robustness of tile-based FRA to run-to-run shifting. For each set of simulation conditions, a broad range of tile sizes (a user-selected software input) were tested. The simulated datasets were then submitted to FRA. Since each hit in the FRA hit list was known a priori as either a true or false positive from the simulation, receiver operating characteristic (ROC) curves could be made and used to quantify the discovery “success” for all combinations of the modeling variables.

Using these results, recommendations for tile size selection and experimental design are provided, and further supported by comparison to the results to previous applications. Finally, extensions of the simulation method to evaluate the robustness of other chemometric methods are investigated.