Development of a rapid and reliable X-ray fluorescence method for protein determination in soybean grains

• Autor(a) principal: Camargo, Rachel Ferraz de
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/64/64135/tde-04092023-145140/
• topic: Análise de alimentos; Chemometrics; Food analysis; Quimiometria; XRF; XRF

X-ray fluorescence spectrometry (XRF) is a technique widely employed for elemental determination. However, the present study pointed to an unconventional direction evaluating the capability of XRF to determine the protein content of soybeans. This was motivated by the perception that soybean might be soon traded based on protein content rather than total grain weight. Additionally, XRF is simply operated, does not require gases or chemicals and the sample preparation and measurements are rapid. The study hypothesizes that sulfur concentration might proxy protein content in soybeans. The research was divided into two parts. Firstly, sample preparation and data acquisition methods were defined and optimized. Briefly, the proposed method consists in (1) coarsely grinding the grains with a household coffee grinder and then (2) scanning the samples for 90 s with the X-ray tube set at 40 kV and 30 A. Employing 108 samples in the calibration set, a logistic regression model was developed to classify soybean into high- or low-protein groups. The model was validated using an independent set of 54 samples. At validation, the global accuracy and kappa index of the model were 0.81 and 0.61, respectively. The numbers indicate that the technique can be used for classifying soybean based on protein content. In the second part of the research, univariate linear regression, multiple linear regression, and partial least squares regression (PLS) models were established to evaluate the feasibility of quantifying the attribute. The models presented reasonable predictive performance (RPD > 1.57) and PLS presented the highest performance (R2 = 0.73) at the validation, suggesting that the XRF technique can be used for rough screening applications. Additionally, samples prepared by mixing soybeans with soybean flours were added in the calibration (22 samples) and validation (10 samples) sets to widen the protein range. The protein content range was 33.8% - 43.9% and changed to 19.2% - 54% after including the mixtures. In this scenario, higher R2 values were obtained (e.g., R2 = 0.89 for PLS), confirming that protein can be predicted from XRF data. The hypothesis that sulfur proxies the protein content in soybeans was confirmed by the present study, since the sulfer emission line was the most important variable for prediction, regardless of the modeling strategy used