Genomic models account for genomic preselection by correctly estimating Mendelian sampling terms of preselected animals



It has been previously shown that in breeding programs that do not use external data, genomic models estimate breeding values of preselected animals without preselection bias and with minimal accuracy loss, as long as genotypes of preselected animals and of their parents are used in the evaluation. The objective of this paper was to show that genomic models account for genomic preselection (GPS) by correctly estimating the Mendelian sampling terms (MSTs) of preselected animals. We simulated a single-trait breeding goal with heritability of 0.1, and 15 recent generations undergoing selection. To select the parents of the next generation from the animals in the most recent generation, we genomically preselected 10% of males and 15% of females in generation 15. We then performed evaluations of the preselected animals with both genomic and pedigree models, both including and excluding records on the preselected animals. We also conducted another set of genomic and pedigree evaluations without preselection, to serve as control. Results showed that both the true and estimated MSTs in the control scenario were on average zero, regardless of whether they were estimated with genomic or pedigree models. With GPS, the average true MST was positive, was correctly estimated by genomic models, and hugely underestimated by pedigree models. Compared to the MSTs estimated by pedigree models, the MSTs estimated by genomic models in both GPS and control scenarios had variances that were closer to the variances of the corresponding true MSTs. We concluded that genomic models indeed correctly estimate Mendelian sampling terms of preselected animals, and that how they are able to account for GPS.