Avaliação genética de poedeiras utilizando modelos de regressão aleatória

Abstract

The objective of this work was to compare random regression models with different orders of adjustment of the Legendre polynomial and to test different structures of the data file in the genetic evaluation of the animals of the Rhode Island Red lineage breed. The records came from the Centro Nacional de Pesquisa de Suínos e Aves of the Empresa Brasileira de Pesquisa Agropecuária (CNPSA/EMBRAPA). Monthly production records of 4,211 matrices were evaluated, from the 22nd to the 61st weeks of bird life. To model the trajectory of egg production, random regression models were used, which differed in the order of adjustment of the Legendre polynomials for the random effects. Additionally, from a complete structure, six sub-files were also evaluated, with different structures and amounts of information, as follow: structure 1 - odd months; structure 2 - even months; structures 3 and 4 - formed by the months that best described the egg production curve (beginning, peak and persistence); and structures 5 and 6 - determined by principal components analysis, which were selected the months of production responsible for explaining most of the genetic variation of the data. The statistical criteria used to choose the model and structure were: Akaike Information (AIC), Bayesian Schwarz Information (BIC), Maximum Likelihood Function Logarithm (LMV), residue, reliability and Spearman's rank correlation. The model PL24, with good estimatives of heritability (0.04 to 0.22) and genetic correlation (0.28 to 0.99) was the model that promoted the best fit for the egg production of the birds under study. The structure four (which considered months 2, 3, 4, 5 and 10) presented greater reliability with the complete structure, indicating that it is possible to reduce the data and make a good genetic evaluation of laying hens of Rhode Island Red lineage breed.