Potential of NIR spectroscopy to predict amygdalin content established by HPLC in intact almonds and classification based on almond bitterness

Abstract

In this study, 360 intact almonds, half sweet and half bitter, were assessed by near-infrared (NIR) spectroscopy to predict amygdalin content (established by high performance liquid chromatography (HPLC)) and by applying partial least squares (PLS) to the spectral data. After optimising amygdalin extraction and chromatographic conditions, the amygdalin contents found by HPLC were not detected or below to 350 mg kg−1 for sweet almonds, and between 14,700 and 50,400 mg kg−1 for bitter almonds. The intact almond spectra resulted in good predictions of amygdalin content with R2p of 0.939 and RMSEP of 0.373. Almonds were correctly classified into sweet and bitter by linear discriminant analysis (LDA), quadratic discriminant analysis (QDA) and PLS-DA, with sensitivity and specificity values higher than 0.94 for evaluation set samples. Based on these results, it can be concluded that NIR spectroscopy is a good non-destructive alternative to be used as an automatic in-line classification system by food industry.