Multi-scale analysis of structural variability of Caryophyllaceae saponins by a simplex machine learning approach

A mass conservation law-based chemometric approach was developed to extract smoothed processes governing inter- and intra-molecular variability of structural diversity in metabolic pools. The approach consisted of a machine-learning method using simplex rule to calculate a complete set of smoothed barycentric molecules from iterated linear combinations between molecular classes (glycosylation classes). An application to four glycosylation levels (GLs) of Caryophyllaceae saponins highlighted aglycone-dependent variations of glycosylations, especially for gypsogenic acid (GA) which showed high 28-glucosylation levels. Quillaic acid (QA) and gypsogenin (Gyp) showed closer variation ranges of GLs, but differed by relationships between glycosylated carbons toward different sugars. Relative GLs of carbons C3 and C28 showed associative (positive), competitive (negative) or independent (unsensitive) trends conditioned by the aglycone type (GA, Gyp) and molecular (total) GLs (the four classes): 28-glucosylation and 28-xylosylation showed negative global trends in Gyp vs GLs-depending trends in QA. Also, relative levels of 3-galactosylation and 3-xylosylation varied by unsensitive ways in Gyp vs positive trends in QA. These preliminary results revealed higher metabolic tensions (competitions) between considered glycosylations in Gyp vs more associative processes in QA. In conclusion, glycosylations of GA and QA were relatively distant whereas Gyp occupied intermediate position.