Création d'équations de calibration universelles en utilisant la standardisation de l'information spectrale: exemple des acides gras

peer reviewedDue to the interest of fatty acids (FA) for the nutritional quality of milk fat and their links to the physiological status of cows, there is an increasing interest to analyse the FA profile of cow milk by mid-infrared spectroscopy (MIR). Indeed, this technology is worldwide used for milk analysis and allows fast and cheap analysis. To build robust MIR equations, a large spectral dataset including samples covering the maximum chemical variability and FA reference values is needed; however this kind of dataset is difficult to obtain mainly due to the cost of reference analysis. Therefore, the sharing of data is a potential solution, but different brands and types of MIR apparatus exist, and therefore different spectral responses and formats, involving difficulties to directly merge all datasets. The objective of this study was to standardize the spectra coming from different apparatus on the same format through a Piece-wise Direct Standardization (PDS) in order to group them into a common database and then to construct robust MIR FA equations. For this, milk samples were collected in Belgium, United Kingdom, Luxembourg, Ireland, France and Germany during 8 years, and measured in different MIR instruments (Foss and Bentley). The reference FA concentrations were measured by gas chromatography. This allowed to group 1,804 samples covering a large FA and spectral variability as well as different geographical areas, breeds, diets and seasons. Calibrations were performed using the chemometric tool Partial Least Squares (PLS) and the first results showed R2 ranging from 0.67 to 0.99 and a ratio of standard error of cross-validation to standard deviation (RPD) ranging from 1.73 and 9.74. The perspective of this work is to allow the use of unique (universal) equations in different countries and different MIR apparatus