Vibrational spectroscopic techniques and variable selection in Linear Discriminant Analysis to geographical origin discrimination of Jatropha mollissima sap

Abstract

This study aimed at the geographical origin discrimination of Jatropha mollissima saps using mid- and near-infrared spectroscopies (MIR and NIR, respectively) and Linear Discriminant Analysis (LDA). For this purpose, a total of 108 sap samples were collected in 3 different geographical regions over 12 months, and their content of total polyphenols, flavonoids, and tannins were quantified. Overall, samples from region C had consistently lower levels of these secondary metabolites throughout most of the months studied. In contrast, samples from regions A and B displayed relatively stable metabolite concentrations over the collection period. Since raw sap samples are subject to deterioration during storage, lyophilization was employed to remove moisture and consequently increase their shelf-life. Thus, MIR spectroscopy was applied to both raw and lyophilized sap samples, while NIR spectroscopy was only applied to lyophilized samples. Then, Principal Component Analysis from the secondary metabolites and spectral data indicated a trend of separation between the region C (located in Sertão) and the regions A and B (located in Agreste). Next, the Successive Projection Algorithm (SPA), Genetic Algorithm (GA), and Ant Colony Optimization (ACO) were used for variable selection in LDA. As a result, all constructed models achieved a sensitivity, specificity, and accuracy of 100 %, or very close to it, in both the training and test sets. However, the SPA-LDA models were more parsimonious, selecting fewer variables and presenting reproductive results as it is a deterministic technique. Hence, the proposed analytical methodologies align with the principles of Green Chemistry by requiring a simple sample preparation, avoiding the use of reagents and solvents, and reducing waste generation. Moreover, special attention can be given to NIR spectroscopy, as it offers a cost-effective analytical tool that can be explored in situ using a portable miniaturized device in the future.