Interim genomic prediction considering newly acquired genotypes and phenotypes


  • Jeremie Vandenplas Wageningen Uinversity & Research
  • Herwin Eding
  • Mario Calus


In the context of single-step genomic evaluations, current methods for predicting interim genomically enhanced breeding values (GEBV) for young genotyped animals only consider new genotypes while ignoring new phenotypes. The aim of this study was to develop a method for predicting interim GEBV for animals associated with genotypes and/or phenotypes not included in a previous single-step genomic evaluation. The method thus developed relies on a Bayesian view of the linear mixed model for pedigree Best Linear Unbiased Prediction (BLUP). Assuming that single nucleotide polymorphism (SNP) effects are known a priori, it can be shown that a single-step genomic BLUP is equivalent to a pedigree BLUP with a prior mean based on direct genomic values and a prior covariance structure matrix requiring only pedigree information. Our method was tested on real data extracted from the December 2019 run of the Dutch-Flemish 4-trait evaluation for temperament and milking speed. The initial single-step evaluation included 6 520 406 animals (including 444 genetic groups), 4 147 302 records, and 144 086 genotypes. Four subsequent monthly single-step evaluations were performed by adding genotypes and phenotypes acquired during the corresponding additional period. Interim GEBVs for all animals in the pedigree were also computed for each month using our method and based on the SNP effects estimated by the initial single-step evaluation. For all traits, Pearson correlations between estimated SNP effects obtained from the initial single-step evaluation and from the four subsequent single-step evaluations decreased slightly over time, but were all higher than 0.98. By considering GEBVs obtained from the subsequent monthly single-step genomic evaluations as the references, accuracies of interim GEBVs for animals that were only newly genotyped, only phenotyped, or both, ranged between 0.98 and 1.00 across all traits. The corresponding dispersion biases ranged between 0.99 and 1.01. Therefore, our method results in accurate interim genomic predictions for all groups of animals.