|
|
|
||
|
|
EUROPEAN HARMONISATION OF NEAR INFRARED REFLECTANCE SPECTROSCOPY (NIRS) ANALYSIS ONRABBIT FEEDS: FIRST STEPS XICCATO, G.(1), TROCINO, A.,(1), MAERTENS, L.,(2) , DE BOEVER, J.L.,(2), PEREZ, J.M.(3), ANDRIEU, J.(4) (1)
Dipartamento di Scienze Zootecniche, Universitā di Padova, Italy
ABSTRACT: The harmonisation of experimental methodologies is the goal of several European projects allowing a more efficient comparison of scientific results among laboratories. Among these projects, the Concerted Action ERAFE (European harmonisation of rabbit feed evaluation; FAIR3-1651) involves six European laboratories working on rabbit nutrition. One of the ERAFE tasks is the harmonisation of methods to predict the nutritive value of feeds through Near Infrared Reflectance Spectroscopy (NIRS). As a first step, the project aimed to compare NIRS accuracy in predicting chemical composition and nutritive value of 55 compound feeds for rabbits scanned in three laboratories (Belgium, France and Italy), following a common procedure of sample preparation. The pre-drying of samples was decided in order to homogenise moisture content, which is one of the most important factors of variation in NIRS analysis. The samples were scanned after drying in a ventilated oven at 60°C overnight. Scanning method and spectra analysis were not harmonised among laboratories. Calibration equations were calculated in all laboratories both on absorbance and transformed data by step-wise multiple linear regression (MLR) and Partial Least Square Regression (PLSR) in order to predict crude protein (%DM), crude fibre (%DM), gross energy (MJ/kg DM), dry matter digestibility (%), gross energy digestibility (%) and digestible energy (MJ/kg DM). Moreover, calibrations were recalculated in one laboratory by using a unique software. The calibrations calculated in the three laboratories did not show large differences concerning the coefficient of determination or accuracy of prediction for all variables. The derivative treatment of spectra permitted to improve the accuracy of calibrations compared to absorbance data. No sensible advantage of MLR or PLSR was evidenced. When the different sets of spectra were submitted to calibration by using a unique software, no difference in comparison with the previous results was obtained. Principal component analysis separated spectral data according to the laboratory. It can be concluded that despite the different hardware and software equipment used, the calibration results for all variables were substantially equivalent in all labs when NIRS analysis was performed with a harmonised sample preparation and following a similar mathematical and statistical procedure for data analysis.
|
|
|
