APPLICATION OF DATA ANALYTICS AND SPECTROSCOPY ON THE CHEESEMAKING PRODUCTION CHAIN

The present thesis aims to provide a rigorous methodology for investigating specific dairy topics that remain somewhat underexplored in the food research. Data analysis techniques were employed to examine the impact of process variables on the food matrix at selected stages of cheesemaking, following the product’s path along its process line. As a first step, the issue of milk coagulation process monitoring through non-intrusive sensors is taken under consideration. Specifically, Raman spectroscopy combined with multivariate statistical methods is employed to assess its ability to keep track of milk gelation dynamics and to predict the curd cutting time, a key technological parameter in dairy processing. The following step is to exploit the whole potential of Raman spectroscopy in identifying the chemical bonds that behave as tracers for the coagulation process. This identification, integrated with appropriate signal preprocessing and the implementation of a kinetic model, enables the development of a monitoring system capable of providing real-time estimates of both concentration of compounds involved in milk coagulation and the reconstruction of curd’s unmeasurable variables such as the elastic modulus, since its measurement requires the invasive extraction of samples. Raman spectroscopy tool has also been then employed for the development of a fault detection system through multivariate statistical process control algorithms. The overall results show that Raman spectroscopy can provide multitasked assistance in the cheese manufacturing process. The subsequent part of this dissertation shifts to the analysis of cheese metabolic profile, in order to study the effect of different process variables on the final product quality. The metabolomic study begins with the analysis of fatty acids and elemental composition of cheese sample produced from milk obtained in three different seasons (Summer, spring and winter). Subsequently, Univariate tests on Gas Chromatography–Mass Spectrometry peaks complemented with false discovery rate correction are employed to identify potential biomarkers for different thermal treatment and ripening time for PDO Fiore Sardo cheese, resulting in the significance of biogenic amines, endocannabinoids, sugars and organic acids. Finally, the differences in the lipidic profile of Pecorino Romano cheese over different rennet sources are here investigated using liquid chromatography-quadrupole time-of-flight tandem mass spectrometry. For the first time, a lipidomic study is conducted to differentiate vegetable-renneted (Cardoon) cheese samples from calf rennet-based product. The interaction between different rennet sources and cheese ripening times is also considered. Due to the high dimensionality and the presence of numerous non-informative variables within the lipidomic dataset, the performance of classification models can be significantly hindered. To address this issue, the statistical procedure proposed in this study foresees a variable selection strategy aimed at improving the classification accuracy of multivariate classifiers. In fact, when distinguishing between different types of rennet, the classifier's accuracy increases markedly (from 56% to 88%) when using selected features instead of the full set. The selected features are considered as biomarkers for distinguishing among the different renneting recipes. The biomarkers span several major lipid classes, including glycerophospholipids, triacylglycerols, monoacylglycerols, sphingolipids, various fatty acids and cholesterols. Overall, the outcomes presented in this dissertation demonstrate that spectroscopic platforms coupled with data analytics tools can be a promising tool within the cheese manufacturing workflow, offering a holistic perspective which considers the production processes as an interconnected system rather than a set of isolated units.