Use of visible-near infrared spectroscopy to predict nutrient composition of poultry excreta

Abstract

Nowadays optimal feed formulation for poultry is sought for available content, which takes into account how the nutrients are digested and metabolized by the animal. The digestibility coefficients of the nutrients are usually obtained in in vivo trials that require feeding the birds with different diets of well-known composition and analyzing a large number of excreta samples. Nutrient excreta composition is usually found by wet analytical methods. This work presents visible-near infrared (Vis-NIR) calibrations for organic matter, protein, fat, gross energy, uric acid and phosphorus in excreta from bioassays involving broiler chickens, laying hens and broiler turkeys carried out between 2017 and 2020. The Vis-NIR spectra (400–2499.5 nm) were pretreated by generalized least squares weighting (GLSW) and partial least squares regression (PLSR) was used to obtain the prediction models. The six parameters were properly predicted with the values of ratio of performance of deviation (RPD) and coefficient of determination of prediction (R2p) of the validation set ranging from 3.7 to 4.6 and from 0.91 to 0.95 respectively. All but one of the calibrations passed the statistical tests for fit for purpose described in ISO 12099:2017. Despite the global calibrations provided satisfactory results, specific calibrations for broiler chicken excreta and for laying hen excreta were developed to check if their predictions could be even better but the results did not improve. Finally, the root mean square error of prediction (RMSEP) of the global calibrations was compared with the standard error of the reference methods employed for the analysis of these parameters, confirming their high performance and direct applicability.