Linkage disequilibrium and genomic selection in pigs

Securing a sufficiently large set of genotypes and phenotypes can be a limiting factor when implementing genomic selection. This limitation may be overcome by combining data from multiple populations or by using information of crossbred animals. The research described in this thesis characterized linkage disequilibrium (LD) patterns in different pig populations and evaluated whether the consistency of LD between populations allows us to make predictions about the performance of genomic selection when multiple populations are included in the prediction and/or validation datasets.

In chapter 2 I evaluated the persistence of LD and patterns of LD decay of pure and crossbred pig populations using real data that was representative of the crossbreeding structure of pig production. The persistence of phase between the crosses and their parental populations was high, indicating that similar marker effects might be expected across these populations. Across the purebred populations the persistence of phase was low therefore higher density panels should be used to have the same marker-QTL associations across these populations.

In chapter 3, the well-known nonlinear model developed by Sved (1971) was compared against a an alternative, loess regression, to describe LD decay. The loess regression model was found to be less influenced by the lack of residual normality, independence and homogeneity of variance than the nonlinear regression model. The loess regression model resulted in more reliable LD predictions and can be used to formally compare the LD decay curves between populations. 

Chapter 4 showed the utility of different reference sets (across- and multi-population) for the prediction of genomic breeding values, as well as the potential of using crossbred performance in genomic prediction. None of the accuracies obtained using across-population, or multi-population genomic prediction, nor the accuracies obtained using crossbred data, followed the expectations based on LD that was described in chapter 2. I showed that across-population prediction accuracy was negligible even when the populations had common breeds in their genetic background. The variable accuracies of multi-population prediction and moderate accuracy of prediction of crossbred performance appeared to be a result of the differences in genetic architecture between pure populations and between purebred and crossbred animals.

 In chapter 5, a methodology that uses information from genome wide association analyses in the genomic predictions was developed and evaluated. The aim in chapter 5 was to let the genomic prediction model use information from the genetic architecture in single- and multi-population genomic prediction. I showed that using weights based on GWAS results from a combined population did result in higher accuracies of GBLUP in single- as well as in multi-population predictions.

In chapter 6 I placed my results in a broader context. I discussed about the theoretical and practical aspects of linkage disequilibrium in breeding and in the estimation of effective population size. I also discussed the application of genomic selection in a small population and in practical pig breeding, including the prospects of using whole genome sequence for genomic prediction.