RESUMO
A rapid spectroscopic method to determine total protein in bovine and buffalo milk using UV spectra of guanidine-hydrochloride mixed milk has previously been reported and validated. The method was based on mixed calibration samples and univariate calibrations of fourth derivative (4D) spectra. In this study the same method was compared and tested for determination of total protein in goat milk. Calculations based on multivariate calibration (partial least squares regression) on full spectra of goat milk were used. The method was tested on 2 UV instruments. The comparison resulted in a significantly more robust (i.e., better) transferability between UV instruments for the partial least squares regression method on full spectra compared with previous univariate calibration of 4D spectra. Local (1 instrument) calibrations gave similar, significantly not different (chi-squared test) cross-validated prediction error results for the 2 methods. It can be concluded that there is no need for fourth derivation. Partial least squares regression on full spectra was equal or superior to using the 4D spectra.
