Exploiting the successive projections algorithm to improve the quantification of chemical constituents and discrimination of botanical origin of Argentinean bee-pollen

Abstract

Bee-pollen as a functional food is gaining importance throughout the world because of its composition and biological properties. The protein content is one of the main parameters to determine its nutritional value, but it makes accurate labeling difficult due its high variability related to the botanical origin. Thus, this work employed near-infrared (NIR) spectroscopy and chemometrics to perform the quality control of Argentinean bee-pollen. Compared to full spectrum models, the successive projections algorithm (SPA) for selection of intervals or individual variables always achieved the best results for quantitative and qualitative approaches. For moisture and total protein content determinations, SPA coupled with partial least squares (iSPA-PLS) and multiple linear regression (SPA-MLR) achieved relative errors of prediction (REP) of 3.53% and 3.93%, respectively. For the pollen classifications, in terms of total protein content (as a dietary supplement with a cut-off higher than 20 g/100 g) and botanical origin, discriminant analysis by iSPA-PLS-DA achieved the best predictive abilities, misclassifying only one sample in the test set for both studies. The overall accuracies were 97.2% and 96.1%, respectively. Therefore, NIR spectroscopy combined with chemometrics can be used as an effective, fast, and low-cost tool for screening the quality of bee-pollen.