Strategies for sourdough bread authentication by NIR spectroscopy considering experimental and commercial samples

This study evaluated the potential of near-infrared (NIR) spectroscopy combined with chemometric to authenticate breads according to the leavening agent used, namely sourdough (SD) versus baker's yeast (BY), considering either crust-wise or crumb-wise acquisition configurations. A comprehensive dataset was collected including experimental breads (n = 84), made to account for the largest possible variability of SD-based bread, and commercial samples (n = 62). Exploration by principal component analysis (PCA) did not reveal a clear clustering between SD and BY breads whereas systematic differences between experimental and commercial samples were observed. Partial least squares-discriminant analysis (PLS-DA) models were therefore developed. In a first approach, PLS-DA calibrated on the sole experimental breads achieved excellent performance within the calibration domain but failed to generalize to commercial samples. Then, a second calibration approach was tested, by combining experimental and commercial samples in a single dataset. PLS-DA models showed more robust discrimination between SD and BY breads, with higher sensitivity than specificity for the SD class. This is desirable given the higher added value of sourdough products and the need for effective protection against misdescription and fraud. Comparable classification performance was obtained for both crumb-wise and crust-wise acquisitions, supporting the applicability of NIR sensors under different practical conditions, including the analysis of sliced or whole bread loaves. These results confirm the suitability of NIR spectroscopy for bread authentication extending previous findings by including a broad range of sourdough formulations and commercial products, while underscoring the necessity of representative calibration sets when modelling complex and heterogeneous food matrices.