Near-infrared spectroscopy to assess chemical composition of sheep and goat cheeses

The present study aimed to evaluate the performances of Fourier transform near-infrared spectroscopy technique to determine the chemical and the fatty acid composition of different types of cheeses. A total of 95 cheeses from sheep and goat raw milk were produced in small local dairies of Siena province (Tuscany). For each cheese, spectrum was collected in intact slices of the sample and fatty acid profile was determined in ground samples. Outliers were identified and different mathematical pre-processing treatments (SNV, MSC, baseline correction and de-trending) were applied when necessary. Considering traditional chemical analysis and raw cheese spectral data, calibration and cross-validation models were carried out using partial least squares regression (PLS). The best results were evaluated in terms of coefficient of determination in calibration and cross-validation (R2cv), and root mean square error in calibration and cross-validation, and residual prediction deviation (RPD). Moisture, protein and ash showed the best R2cv (0.89, 0.74 and 0.72, respectively) and RPD values (3.0, 2.6 and 2.1, respectively). Saturated, monounsaturated and polyunsaturated fatty acids showed R2cv which ranged from 0.75 to 0.67, and RPD <2.0. Intermediate results in terms of R2cv (0.62 as mean) were obtained for medium chain saturated fatty acids (C8:0 to C14:0), whereas for C18 series only oleic acid reached good accuracy of prediction (R2cv >0.70). Obtained results are promising and additional samples could strongly increase the predictive ability for small dairy farms.