Advanced chemometrics to predict cheese-making traits from milk infrared spectra

Predicting cheese-making traits from milk spectra represents a significant opportunity for improving the efficiency of the dairy industry. This thesis explores advanced chemometric approaches to predict cheese yield (%CY) and nutrient recovery (%REC) traits using Fourier-transform infrared (FTIR) spectroscopy of milk samples. Through three interconnected studies, we investigated various aspects of this prediction challenge: i) developing a novel Targeted Interval Partial Least Squares (iPLS) approach using spectra from individual milk samples from Brown Swiss cow; ii) using Bayesian models on spectra from bulk milk samples from Grana Padano PDO production and iii) examining the impact of the production system within the Parmigiano Reggiano PDO consortia on the prediction accuracy of cheese-making efficiency. Our findings demonstrated that the use of milk spectra, combined with appropriate chemometric techniques, can effectively be used to predict cheese-making traits. Overall, these findings showed that the selection of spectral regions associated with the most important components of milk enormously improve the accuracy of prediction, suggesting the potential for developing specialized instruments with custom calibrations suitable also for online prediction. Also, valuable insights have been provided regards to the implementation of spectral prediction methods in industrial cheese-making processes, offering a pathway to improve production efficiency through rapid, non-destructive measurements, as well as the possibility to understand the impact of different factors related to the individual animal, environment, and cheese-making procedures directly on the predictability of the cheese-making traits.