Discrimination of intact almonds according to their bitterness and prediction of amygdalin concentration by Fourier transform infrared spectroscopy

Intact almond kernels (N = 360, half sweet and half bitter) were analyzed using attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) for the prediction of amygdalin concentration and to classify them according to their bitterness. Amygdalin concentrations for sweet and bitter almonds, determined by high performance liquid chromatography, were between 0.7-350 and 15,000-50,000 mg kg(-1), respectively. Concentrations were successfully predicted by applying partial least squares (PLS) to the pre-treated spectral data with R-p2 of 0.951 and RMSEP of 0.398. Additionally, linear discriminant analysis (LDA), quadratic discriminant analysis (QDA) and PLS-DA models were constructed to classify samples according to their bitterness. All three models provided a satisfactory discrimination of almonds into sweet and bitter categories, providing overall accuracy values of 83.3%, 86.1% and 98.6%, respectively. The results indicate the potential of ATR-FTIR spectroscopy for the reliable, easy and fast prediction of amygdalin concentration, and for almond classification according to their bitterness.