Speaker
Description
This study investigates the prediction reliability of cheese making traits [3 measures of cheese yield (%CY): fresh, solids and retained water (%CYCURD, %CYSOLIDS, %CYWATER); 4 recovery traits (%REC): milk fat, protein, solids and energy in the curd (%RECFAT, %RECPROTEIN, %RECSOLIDS and %RECENERGY)] (Ferragina et al., 2013) applying Bayesian models on the Fourier-transform infrared spectroscopy (FTIR) spectra of vat milk samples used for Grana Padano PDO production. Information from 50 cheese-making days (in total 139 vats) from two dairy industries were collected. For each vat, the milk, and the cheese after 48h from cheese making, were weighted while milk and whey composition (total solids, lactose, protein and fat) was analyzed. Two spectra from each milk sample were collected in the range between 5,011 and 925 cm-1 and averaged prior the data analysis. The calibration models were developed with a Bayesian approach (Ferragina et al., 2015) by using the BGLR (Bayesian Generalized Linear Regression) package of R software (R Core Team, 2013). Performance of models was assessed by coefficient of determination (R2VAL) and the root mean squared error of validation (RMSEVAL). A random cross-validation (CV) was applied [80% calibration (CAL) and 20% validation (VAL) set] with 10 replicates. The most accurate predictions after CV were obtained for %CYCURD and %CYSOLIDS, which exhibited R2VAL and RMSEVAL values of 0.55 and 0.27, and of 0.65 and 0.18, respectively. The %CYWATER showed the lowest R2VAL (0.53), being the least repeatable among cheese making traits. Considering %REC traits, promising results were obtained for the recovery of protein (RMSEVAL = 0.31%). In opposite, the recovery of energy (RMSEVAL = 0.81%) and fat (RMSEVAL= 1.83%) showed a less favorable result. These results demonstrate FTIR spectroscopy could be a valid method to indirect monitor cheese making efficiency at the dairy industry level for Grana Padano PDO production.
