GC-MS characterisation of novel artichoke (Cynara scolymus) pectic-oligosaccharides mixtures by the application of machine learning algorithms and competitive fragmentation modelling

Carbohydr Polym. 2019 Feb 1:205:513-523. doi: 10.1016/j.carbpol.2018.10.054. Epub 2018 Oct 24.

Abstract

Novel artichoke pectic-oligosaccharides (POS) mixtures have been obtained by enzymatic hydrolysis using four commercial enzyme preparations: Glucanex®200G, Pentopan®Mono-BG, Pectinex®Ultra-Olio and Cellulase from Aspergillus niger. Analysis by HPAEC-PAD showed that Cellulase from A. niger produced the greatest amount of POS (310.6 mg g-1 pectin), while the lowest amount was produced by Pentopan®Mono-BG (45.7 mg g-1 pectin). To determine structural differences depending on the origin of the enzyme, GC-MS spectra of di- and trisaccharides have been studied employing three machine learning algorithms: multilayer perceptron, random forest and boosted logistic regression. Machine learning models allowed characteristic m/z ions patterns to be established for each enzyme based on their GC-MS spectra with high prediction rates (above 95% on the test set). Possible chemical structures were given for some m/z ions having a decisive influence on these classifications. Finally, it was observed that several ions could be formed from specific POS structures.

Keywords: Artichoke pectin; Enzymatic hydrolysis; In silico fragmentation; Neural network; Pectic-oligosaccharides.