GC‐MS characterisation of novel pectic‐oligosaccharides derived from artichoke pectin using machine learning and competitive fragmentation modelling (CFM‐ID)

Abstract

Pectic oligosaccharides (POS) comprise a wide variety of structures containing different monomers including galacturonic acid (GalA), rhamnose (Rha), xylose (Xyl), arabinose (Ara) or galactose (Gal). To characterise these mixtures, several chromatographic techniques might be employed. GC‐MS‐EI is widely used in the field of metabolomics due to its reproducibility and simplicity. It generates massive amounts of high‐dimensional data and standards of many oligosaccharides might not be commercially available. To interpret GC‐MS‐EI results, in silico fragmentation methods such as CFM‐ID have been developed. On the other hand, machine learning allows extracting chemically relevant information from complex GC‐MS‐EI spectra. Therefore, the purpose of this study was to characterise novel POS obtained from hydrolysis of artichoke pectin using different types of enzymes. With this aim, a GC‐MS‐EI data analysis strategy based on the combination of machine learning and in silico fragmentation to interpret these models is presented. First, POS were obtained from pectin using different glycosidases (Pectinex Olio, PO; Cellulase from Aspergillus niger, CEL; Pentopan, PEN) and then reaction mixtures were analysed by GC‐MS‐EI. Spectra of di‐ and tri‐POS were classified by determining structural patterns depending on the enzyme selected. Models used like random forests and artificial neural networks led to high prediciton rates, showing accuracies over 95% on clasification of new samples. After, chemical structures of most influent m/z ions were calculated using an oligosaccharide in silico fragmentation library previously generated using CFM‐ID. Some relevant ions in POS obtained with PO (m/z 340, 424 and 443) were originated from Rha‐ɑ(1,4)‐GalA, GalA‐ɑ(1,2)‐Rha or Xyl‐ɑ(1,3)‐GalA (methylated/acetylated or not). Similarly, POS obtained with CEL showed high abundances of m/z 482, derived from GalAɑ(1,2)‐Rha (acetylated) and Xyl‐ɑ(1,3)‐GalA (acetylated). In contrast, POS obtained with PEN showed high abundances of m/z 454 which could also be formed from POS containing Araɑ(1,4)‐Rha. In conclusion, this methodology allows to tentatively predict some of the most probable POS structures that may be obtained from artichoke pectin using different enzymes.