Deliverable D5.3 New methodologies to account for repeated measurements, social effects, and variability in performance in genetic evaluations

Objectives

The general objective of this task was to develop statistical models and procedures as well as new software for selection on feed efficiency and robustness. Specifically we have worked on three main topics of research:

1. Improvement of models and procedures for selection on feed efficiency taking into account the characteristics and structure of the data
2. Definition of models and procedures to account for indirect genetic (IGE) and environmental effects (i.e., social interaction effects) on growth and feed efficiency
3. Selection for the individual’s environmental sensitivity (robustness) and interaction between the genotype and feeding regimen

Rationale

Pig and rabbit data originating from task 5.1 and task 5.2 were used to implement the proposed models and procedures in task 5.3, which were developed following a frequentist or a Bayesian approach. The implementation of the statistical methodologies were developed from existing software and their details are available in the published papers. Models were compared in terms of goodness of fit, predictive ability, and response to selection on real and simulated data. In addition, a freely available software module was developed that allows multiple-trait analyses.
Regarding the first topic of research, we have developed and implemented the structured antedependence model (SAD) for the analysis of longitudinal records of feed conversion ratio (FCR). This model fits the covariance structure of the data with few parameters better than other models and provides unbiased estimates of the correlation between distant measurements of the trait. A selection criterion combining the weekly breeding values using the eigen-decomposition of the genetic covariance matrix was proposed to summarize the individual estimated breeding value (EBV) trajectory into a reduced number of parameters for practical applications to selection. We have also assessed the impact of missing records of body weight on the prediction of breeding values and estimation of genetic parameters of FCR and have proposed a procedure to impute them. The potential of genomic information to improve the accuracy of breeding value predictions for residual feed intake, applying single step genomic approaches to the random regression and SAD models was also assessed. We proposed a procedure to estimate breeding values and genetic and environmental parameters of feed efficiency from group records of feed intake and individual records of body weight gain and metabolic weight at fattening. In this procedure, feed efficiency is defined as a conditional trait derived from elements of genetic and environmental covariance matrices. These measurements of feed efficiency are genetically and phenotypically uncorrelated with production traits. Thus, selection for this trait does not lead to unfavourable correlated responses on production traits.

Models and procedures to account for the genetic and environmental social interaction effects (also known as indirect genetic effects, IGE) on growth and feed efficiency were proposed. A specific degree of interaction between each pair of animals sharing the same pen/cage was defined by the correlation between each pair of pen/cage mates for different feeding behaviour variables obtained from electronic feeder data. This information alleviates collinearity and improves the model performance and led to a more accurate genetic evaluation of traits affected by IGE. We also have investigated how IGEs vary over time implementing a structured antedependence (SAD) model that includes IGEs. We have evaluated, by simulation, the response to selection on longitudinal average daily gain (ADG) using different selection strategies. We have also estimated the effect of the interaction between the genotype and feeding regimen and we have elucidated the origin of such an interaction.

Finally, we have developed a software module for genetic analysis of multiple-traits heteroscedastic models. This software allows to estimate the magnitude of animal genetic sensitivity to express a specific trait in different environments or for different traits. Thus, it allows knowing the genetic determinism of a definition of global robustness as the genetic correlation between the sensitivities of different traits to environmental variations. The software is implemented in ASReml-4.