Analysis of egg production in layer chickens using a random regression model with genomic relationships

• Published in: Poultry science Vol. 92; no. 6; pp. 1486 - 1491
• Main Authors: Wolc, A, Arango, J, Settar, P, Fulton, J E, O'Sullivan, N P, Preisinger, R, Fernando, R, Garrick, D J, Dekkers, J C M
• Format: Journal Article
• Language: English
• Published: England 01.06.2013
• Subjects: Animals; Chickens - genetics; Chickens - physiology; Female; Genomics - methods; Models, Biological; Oviposition - genetics; Oviposition - physiology; Time Factors
• ISSN: 0032-5791
• DOI: 10.3382/ps.2012-02882

Random regression models allow for analysis of longitudinal data, which together with the use of genomic information are expected to increase accuracy of selection, when compared with analyzing average or total production with pedigree information. The objective of this study was to estimate variance components for egg production over time in a commercial brown egg layer population using genomic relationship information. A random regression reduced animal model with a marker-based relationship matrix was used to estimate genomic breeding values of 3,908 genotyped animals from 6 generations. The first 5 generations were used for training, and predictions were validated in generation 6. Daily egg production up to 46 wk in lay was accumulated into 85,462 biweekly (every 2 wk) records for training, of which 17,570 were recorded on genotyped hens and the remaining on their nongenotyped progeny. The effect of adding additional egg production data of 2,167 nongenotyped sibs of selection candidates [16,037 biweekly (every 2 wk) records] to the training data was also investigated. The model included a 5th order Legendre polynomial nested within hatch-week as fixed effects and random terms for coefficients of quadratic polynomials for genetic and permanent environmental components. Residual variance was assumed heterogeneous among 2-wk periods. Models using pedigree and genomic relationships were compared. Estimates of residual variance were very similar under both models, but the model with genomic relationships resulted in a larger estimate of genetic variance. Heritability estimates increased with age up to mid production and decreased afterward, resulting in an average heritability of 0.20 and 0.33 for pedigree and genomic models. Prediction of total egg number was more accurate with the genomic than with the pedigree-based random regression model (correlation in validation 0.26 vs. 0.16). The genomic model outperformed the pedigree model in most of the 2-wk periods. Thus, results of this study show that random regression reduced animal models can be used in breeding programs using genomic information and can result in substantial improvements in the accuracy of selection for trajectory traits.