Symposium review: Single-step genomic evaluations in dairy cattle

摘要

During the last decade, genomic selection has revolutionizeddairy cattle breeding. For example, Nordicdairy cows (Denmark, Finland, and Sweden) born in2018 were >90% sired by young genomically testedbulls. Thus, the average age of sires for Red DairyCattle cows born in 2018 was only 3.1 yr, whereas in2011 it was 5.7 yr. Earlier the key driver of geneticprogress was the selection of progeny-tested sires,but now it is the genomic preselection of young sires.This leads to a biased estimation of genetic progressby the traditional genetic evaluations. When these areused as input for multi-step genomic evaluations alsothey became distorted. The only long-term solutionto maintain unbiasedness is to include the genomicinformation in evaluations. Although means for singlestepevaluation models were introduced in 2010, theyhave not yet been implemented in large-scale nationaldairy evaluations. At first, single-step evaluations werehindered by computational cost. This has been largelysolved, either by sparse presentations of the inversesof the genomic relationship (G) and pedigree relationship(A_(22)) matrices of genotyped animals needed in thesingle-step evaluation models based on G (ssGBLUP),or by using the single-step marker models. Approachesfor G~（−1） are the APY-G, where the relationships among“young” animals are completely determined by their relationshipto the “core” animals, and single-step evaluationswhere the G~（−1） is replaced by a computationalformula based on the structure of G (ssGTBLUP). Thesingle-step marker models include the marker effectseither directly, as effects in the statistical model, orindirectly, to generate genomic relationships amonggenotyped animals. Concurrently with developmentof the algorithm, computing resources have evolved inboth availability of computer memory and speed. Theproblems actively studied now are the same for bothof the single-step approaches (GBLUP and markermodels). Convergence in iterative solving seems to getworse with an increasing number of genotypes. Theseproblems are more pronounced with low-heritabilitytraits and in multi-trait models with high genetic correlationsamong traits. Problems are also related to theunbalancedness of pedigrees and diverse genetic groups.In many cases, the problem can be solved by properlyaccounting for contributions of the genotyped animalsto genetic groups. The standard solving approach ispreconditioned conjugate gradient iteration, in whichthe convergence has been improved by better preconditioningmatrices. Another difficulty to be considered isinflation in genomic evaluations of candidate animals;genomic models seem to overvalue the genomic information.The problem is usually smaller in single-stepevaluations than in multi-step evaluations but is moredifficult to mitigate by ad hoc adjustments.