Energia metabolizável de alimentos para frangos de corte determinados in vivo e por métodos indiretos

Abstract

The objective of this study was to update the database that generated the AMEn Predictor application, obtain new AMEn Prediction models and, subsequently, validate the application and the updated indirect methods to predict nitrogen-corrected apparent metabolizable energy (AMEn) values of corn (C), wheat bran (WB), soybean meal (SBM) and meat and bone meal (MBM), used in broiler diets. To update the database, a bibliographic review of national and international studies was carried out, from 1986 to 2023, to catalog information on AMEn values and chemical composition (crude protein; ether extract; mineral matter; crude fiber, neutral detergent fiber, acid detergent fiber, calcium and phosphorus) of foods of plant and animal origin, seeking to obtain AMEn prediction models of these foods, using their chemical composition. The correlations between the chemical components of the feedstuffs and AMEn were verified and, subsequently, multiple linear regression (MLR) equations and artificial neural network (ANN) models were determined for each food category (energy and protein), through meta-analysis. To select the best MLR models, the Akaike criterion obtained by the stepwise procedure was used and, for the ANN, a simultaneous statistical analyses of suitability criteria was performed for the training, test and validation sets for each implemented network. For validation, two metabolism experiments (in vivo) were conducted with broilers in growth (15 to 23 days of age) and in the final phase of rearing (36 to 42 days of age). The AMEn of samples of C, WB, SBM, MBM, and mixture between the FCOs were determined. Also, laboratory analyses for determination of chemical composition of the tested feedstuffs were performed, which were utilized in calculating AMEn by the proposed indirect methods. The statistical analysis was made by adjustment of simple linear regression of observed in vivo values on the estimated values by each indirect method. For plant-based foods, only the RLM equations did not satisfactorily estimate the AMEn values of the test feedstuffs. For animal-based foods, all RLM equations were effective in estimating the AMEn of the test foods. It was concluded that the AMEn Predictor and ANN (MLP 10-5-3-1) can be used to satisfactorily estimate the AMEn of plant-based foods, and the equation AMEn = 3172.5776 + 57.7278EE – 45.5500MM + 8.0873Ca – 31.4115P was the most suitable for estimating the energy value of animal-based foods used for broilers.