Mazivila SJ, et al. Talanta 2020.
The present short communication reports a promising analytical method for authentication of milk based on first-order near-infrared (NIR) spectroscopic data coupled to data driven soft independent modeling of class analogy (DD-SIMCA). This one-class classifier was able to correctly classify all samples of genuine milk powder as members of the target class from samples of milk powder adulterated with melamine and sucrose in a concentration range of 0.8-2% (w/w) and 1-3% (w/w), respectively.
Multivariate curve resolution - alternating least-squares (MCR-ALS) was applied as a complementary chemometric model to DD-SIMCA aimed at retrieving pure profiles, allowing to identify the chemical composition of samples properly attributed in the target class or not, providing further investigation from forensic point of view. In order to extend the prime focus of the present report, which was aimed at developing an appropriate chemometric model for authentication purposes, the quantification analysis was also performed. This was done by successful bilinear data decomposition of NIR spectra into pure profiles for the contributing components contained in the system studied (milk and adulterants), allowing to quantify analytes with strong overlapping profiles, even in the presence of an uncalibrated interferent, as demonstrated in this short communication using MCR-ALS under various constraints in order to decrease the rotational ambiguity.