Xu, L., Yan, S., Cai, C. et al. Food Anal. Methods (2013) 6: 1568. doi:10.1007/s12161-013-9575-y
This paper aimed at developing a nondestructive and rapid method to detect adulterations in Chinese glutinous rice flour (GRF) using near-infrared (NIR) spectroscopy and chemometrics. Because various known and unknown ingredients can be potentially used for food adulteration, the commonly used targeted analytical methods focused on detecting one or more known/suspected adulterants usually cannot catch up with the constant “updating” of new adulterants. Therefore, this paper attempted to achieve untargeted detection by modeling the NIR spectra of pure GRF and analyzing those of test samples. Soft independent modeling of class analogy (SIMCA) and a recently suggested one-class partial least squares (OCPLS) was used to develop class models of pure GRF. To highlight the slight variations in NIR spectra caused by low-level doping and enhance the specificity for detecting extraneous adulterants, unwanted variations in pure GRF spectra should be removed. Smoothing, taking second-order derivative (D2), standard normal variate (SNV), and D2-SNV were performed to improve the raw spectra. One hundred thirty pure GRF samples from six main producing areas were prepared and used for training class models. To validate the specificity of class models, 215 adulterated GRF samples were prepared by blending the pure objects with different levels (1, 2, 4, 8, and 10 % (w/w)) of wheat flour, non-GRF, and an illegal food additive, talcum powder, which have been frequently used for GRF adulteration. The best OCPLS model was obtained with D2 spectra with prediction sensitivity of 1.000 and specificity of 0.916; SIMCA with D2-SNV obtained prediction sensitivity of 1.000 and specificity of 0.902. It was demonstrated that adulterations of GRF with 2 % or higher levels of wheat flour, non-GRF, and talcum powder can be safely detected with D2, SNV, or D2-SNV spectra. The analysis results indicate the specificity of untargeted detection of the three adulterants in GRF can be improved by removing the unwanted within-class variations.