A multiway approach to overcome the peak resolution challenges in gas-chromatography coupled to ion mobility spectrometry

Background: Gas-chromatography coupled to ion mobility spectrometry (GC-IMS) is a powerful platform for untargeted analysis of volatile compounds, combining high sensitivity and short analysis times. However, its practical implementation still presents several analytical and computational challenges. From the instrumental perspective, the high sensitivity of isothermal GC–IMS makes it prone to signal saturation and memory effects, which can compromise the reproducibility of measurements. On the data-analysis side, the nature of GC–IMS outputs produce large and collinear datasets, in which baseline drift, shift in ion mobility and retention time modes, and peak overlap hinder the reliable extraction of relevant chemical information. Results: This work proposes a comprehensive workflow from the analytical strategy to the data analysis using as benchmark a data set of honey samples of different origin. Analysing the GC-IMS data, obtained with an instrument operating in isothermal condition, a prominent and marked memory effect was observed; it could be contained only through the systematic use of frequent blanks. Notably, most papers in the literature do not adopt blanks at such frequency. Attention was then devoted to the preprocessing, to improve data quality prior to modelling while limiting the number of preprocessing steps. A multiway approach, i.e. Parallel Factor Analysis 2 (PARAFAC2) applied interval wise, was adopted, for the first time, to resolve overlapping peaks and extract chemically meaningful components from the GC-IMS landscape. The results obtained were also compared with Multivariate Curve Resolution–Alternating Least Squares (MCR–ALS) and recent improved PARAFAC2 algorithms. Significance: A complete pipeline for the GC-IMS analysis was carefully developed and validated. PARAFAC2, was successfully applied for the deconvolution of GC-IMS data. This approach provided comparable results with previous methods and indicated that the same grade of automation reached in the analysis of GC-MS data with the same approach is feasible.