Comparison of genomic selection approaches for small breeds


  • Chris HOZE INRA, UMR 1313, GABI, 78350 Jouy-en-Josas, France UNCEIA, 149 rue de Bercy, 75012 Paris, France
  • S. Fritz
  • D. Boichard
  • V. Ducrocq
  • P. Croiseau


genomic selection, genomic evaluation, multi-breed evaluation, high-density genotype


Within breed genomic selection based on medium SNP density (50K chip) is now routinely implemented in a number of large cattle breeds. However, building large enough reference populations remains a major challenge for many medium or small breeds. The high density BovineHD BeadChip® (HD) containing 777 609 single nucleotide polymorphism (SNP) developed in 2010 is characterized by short distance linkage disequilibrium expected to be maintained across breeds. Therefore, combined multi-breed reference populations can be envisioned. In France where genomic evaluations are only implemented in the three main dairy breeds, a HD-reference of 1869 animals from these 3 breeds was built. Then, 29 091 50K-genotypes from national genomic evaluation were imputed to high density to form a large HD-reference population. This population was used to develop a multi-breed genomic evaluation and compare genomic selection strategies for small breeds.

In this study, we chose to use a large breed (the Normande breed) to mimic a small breed in order to have a large enough validation population and better compare genomic selection approaches. Three training sets containing respectively 1597, 404 and 194 bulls and a unique validation dataset of 394 animals (the youngest Normande bulls) were formed. For each training set, three approaches were compared: pedigree-based BLUP, within-breed BayesCpi and multi-breed BayesCpi in which the reference population was formed by the Normande training dataset and 4989 Holstein and 1788 Montbéliarde bulls.

We computed the correlations between observed and predicted daughter yield deviations (DYD) for six traits and the different approaches. Compared to pedigree-based BLUP, genomic selection approaches provided an average gain in correlation ranging from 6.7 to 7.6% with the smallest reference population and up to 20% with the largest reference population. Multi-breed genomic selection gave the best results in all situations with an average gain in correlation of 3% compared to within-breed genomic selection. However, the increase in correlation is limited when the within-breed reference population is already large and the achieved accuracies are clearly higher. The slope of regression was closer to one when the number of individual in the reference population increased and was similar between multi-breed genomic evaluations and within-breed genomic evaluations. These results showed that multi-breed genomic selection can be an appealing strategy for small breeds.