Near infrared reflectance spectroscopy (NIRS) was used to predict the goat milk fatty acid (FA) profile. The ability of cow milk broad-based calibration equations to predict the goat milk FA profile was assessed. Three hundred twenty-eight samples in the calibration set and 108 in the validation set were analyzed. We showed that the bias and unexplained error were significant for most of the FA despite an adequate standardized Mahalanobis distance (index to establish the boundaries of a population of samples) which allowed us to test the ability of bovine models to predict goat milk FA profiles. To better predict the goat milk FA composition, a specific goat model was investigated. The cross-validation coefficient of determination (R2CV; proportion of variance explained by the model in cross-validation) and residual predictive deviation (RPD; index which allows to standardize the standard error of prediction (SEP)) were >0.90 and 3, for saturated, monounsaturated, and unsaturated FA, total trans FA, isomer cis9trans11 of CLA, cis9-, trans10-, and trans11-C18:1, respectively. Monitoring of the equation performance of milk FA included the calculation of the bias and unexplained error.
Practical applications: In this work, we evaluated the ability of NIRS to predict FA composition of goat milk and tested the ability of using broad-based cow milk calibration equations to monitor an independent set of goat milk samples. We tested the hypothesis that the calibration equations obtained for FA from cow milk could be applied on goat milk. As the spectra of goat milk are similar to those of cow milk when they are measured by the standardized Mahalanobis distance (index to establish the boundaries of a population of samples), we can accept the hypothesis. It will be possible to use the existing calibration equations for predicting FA profile on spectra of milk of a different specie. However, the rejection of the hypothesis would put in doubt the use of Mahalanobis distance as the unique index for testing the performance of calibration equations on different milk populations for predicting FA.