Detecção de adulteração em farinhas de teff usando espectroscopias NIR e EDXRF em conjunto com quimiometria

Abstract

The cereal Eragrostis tef (Zucc.), known as teff, gained worldwide attention because it is glutenfree and rich in minerals, mainly iron. Because of this, teff flour is attractive to frauds, highlighting the addition of cheaper cereal flours. In this context, the study of fast and inexpensive techniques such as NIR (near-infrared) and EDXRF (energy scattering X-ray fluorescence) spectroscopy are important to identify this type of adulteration. This work evaluated the potential of the NIR and EDXRF methods to identify teff flours adulterated with rice, whole wheat, oat, and rye flours. Two teff crops were adulterated following a {5,4} simplex-lattice experimental design, varying the proportion of teff from 65% to 100% and adulterants from 0 to 35%. The NIR spectra were pre-treated with multiplicative scattering correction (MSC), 2nd derivative or combination (MSC+2nd), and analyzed by principal components (PCA). The 15 kV and 50 kV EDXRF spectra were smoothed and pretreated with MSC, 1st derivative, or combination. Then, the data from the 15 kV and 50 kV conditions were fused in a single set by the common dimension analysis (ComDim), which allows the multiblock exploratory analysis. Partial least squares regression (PLSR) models were constructed to relate the NIR spectra to the teff content in the samples, and multiple linear regression (MLR) models were built to link the ComDim-EDXRF scores. Kennard-Stone and SPXY methodologies were used to separate the calibration and test sets. The number of latent variables (LV) and common dimensions (CD) were defined using 10-fold and 8-fold crossvalidation, respectively. Due to the ash and carbohydrate contents, the PCA of the MSC+2nd spectra showed a clear separation between the adulterant flours and the teff crops. Based on the figures of merit and LV number, the best PLSR model was obtained with the MSC+2nd spectra and SPXY sampling, which used 4 LV and presented R2prediction = 92%, RMSEP < RMSEC (0.0209 and 0.0299) which suggests no overfitting of the data, and RPDprediction > 3 indicating an excellent quantitative predictive capacity of the model. For EDXRF spectra, the best MLR model used 6 CD and MSC-SPXY data, presenting R2prediction equivalent to the NIR model, lower RMSEP (0.0181), absence of overfitting, and quantitative application. The high robustness of the NIR and EDXRF models are highlighted, given the prediction of the percentage of adulteration in the teff flour, even with the differences between teff A and B indicated by the exploratory data analysis. Therefore, NIR and EDXRF spectroscopies with chemometrics proved to be efficient methods for controlling the adulteration of teff flours.