Using breed-origin-of-alleles in genomic prediction

Breeding is based on selecting the genetically best individuals as parents of the next generation. Traditionally, selection relied on breeding values estimated from phenotypic records of the individual itself and its relatives, connected using pedigree information. In the last decade, many livestock and plant breeding programs have adopted genomic selection, which estimates breeding values based on genome-wide marker data. Genomic prediction enables accurate selection early in life, leading to faster genetic improvement.

In pig and poultry breeding the production individuals are crosses between three or four breeding lines. Similar strategies are used in crops like maize, where the breeding lines are intentionally inbred, while in livestock breeding lines inbreeding is restricted. In all these cases, the breeding goal is to improve crossbred performance. Selection in pigs and poultry, however, takes place among individuals within a breeding line, and selection decisions are often based on purebred performance. The genetic correlation between purebred and crossbred performance is generally lower than unity, which can limit the progress in crossbred performance. This limitation can be removed by using a crossbred reference population in genomic prediction. Crossbred individuals carry alleles from several parental lines, which may affect genomic prediction.

We developed a method to derive the Breed-Origin-of-Alleles (BOA) for individuals resulting from crossing non-inbred breeding lines, based on phased genotype data and haplotype frequencies observed in the parental lines. We were able to derive BOA for 92-95% of the alleles in crossbred broilers and pigs. We hypothesized that using BOA in genomic prediction models would increase genomic prediction accuracy for crossbred performance. Empirical accuracies, however, were at best slightly improved, despite the observation that allele frequencies and estimated allelic effects for crossbred performance did differ among the parental lines. Using BOA, relationships with respect to a focal breeding line are only based on alleles from this line, while alleles from other lines are ignored. The limited benefit in accuracy when using BOA is likely because alleles from other lines contribute little to variation in relationships of the focal line. Nevertheless, the BOA methodology is potentially beneficial in genomic prediction in admixed populations in other species, especially if founder populations are very different.

The most likely explanation for this limited benefit is that the alleles from other breeding lines had little impact on variation in genomic relationships with respect to the breeding line of interest, so that considering BOA hardly affected the genomic estimated breeding values.