Estimation & Prediction

Genomic evaluations combining multiple data sources but without sharing genotypes

---

Abstract

Single-step genomic evaluation models combine genomic, pedigree and performance information. The various commonly-used single-step algorithms (ssGBLUP, ssMEM) implicitly or explicitly partition breeding values into marker-based components and residual polygenic effects. Given the Direct Genomic Values (DGVs) which are the marker-based component of the predictions of breeding values on genotyped animals, an alternative formulation is a multivariate model. That model assumes DGVs are correlated with the breeding values for the performance traits. This work presents such a multivariate representation that treats Direct Genomic Values (DGVs) from one or more single-step evaluation as correlated traits alongside performance traits, providing a computationally efficient alternative to explicit single-step implementations. It might be useful when an evaluation is based on admixed data, but more than one source of DGVs is available from different reference populations reflecting the different purebred populations represented in the data. It might also be useful for international evaluations where each country has its own reference population, or for other circumstances where parties are not prepared to share their genotypes, or their marker effects, but are willing to share the DGVs. For genotyped animals, DGVs are treated as observed phenotypes with near-perfect heritability; for non-genotyped animals, predictions of DGVs are imputed using pedigree relationships from DGVs on relatives and from correlated trait information. The multivariate mixed model equations show that predictions of phenotypic trait breeding values are computed in a similar manner to those from single-step models when DGVs are appropriately scaled, and the relevant variance-covariance matrices are used. An important consideration is ensuring the DGVs are on the appropriate scale. Critical parameters include the genetic correlations between DGVs from different sources, and the genetic correlations between DGVs from each source and the phenotypic information. The multivariate framework accounts for correlations among DGV sources, preventing double-counting and allowing the animals to have DGVs from one, multiple, or no sources. This framework provides a practical solution for national genetic evaluations where multiple parties contribute genomic predictions without sharing underlying genotype data.

Keywords: 2026

How to Cite:

Garrick, D. & Garrick, D., (2026) “Genomic evaluations combining multiple data sources but without sharing genotypes”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2287416. doi: https://doi.org/10.31274/wcgalp.24291

Rights: 1