Abstract :
[en] The objective of this study was to estimate if the use of milk prediction equations based on mid-infrared (MIR) spectrum could be extended to other breeds than Holstein (HOL). MIR-based equations are most of the time developed on HOL datasets which can prevent their use for other breeds. However, if the milk spectral variability of other breeds is included in the HOL calibration dataset, the spectral extrapolation is limited. We compared spectral variability between Walloon (Southern Belgium) HOL (n = 6,867,134 spectra) and two other Belgian breeds: the Dual-Purpose Blue (DPB, n = 292,106 spectra) and the East Belgian Red and White (EBRW, n = 42,418 spectra), with records from 2007 to 2023. To achieve this purpose, for each breed, we projected the standardized spectra on the first three principal components (PC) estimated from a sample of one million Walloon spectra. Localization index of each spectrum, computed from the scores of the first three PC, were used to select representative spectra: n = 165,507 for HOL, n = 53,561 for DPB and n = 21,752 for EBRW. Finally, to estimate if the spectral variability of the DPB and the EBRW was included in the spectral variability of HOL, we used PC scores to compute the barycenter of selected HOL spectra weighted by their density. We then computed the GH distance of each DPB and EBRW selected spectrum to this HOL barycenter. If the GH distance of the DPB or EBRW spectrum was higher than 3, this spectrum was considered as outside the spectral variability of HOL. It was found that 98.85% of DPB and 99.53% of EBRW selected spectra were included in the spectral variability of HOL. To conclude, the results suggest that milk MIR prediction equations developed on HOL spectra can be applied to DPB and EBRW without spectral extrapolation.
