Single-step genomic predictions for crossbred Holstein and Jersey cattle using metafounders

The study examined the impact of incorporating metafounders (MF) in single-step GBLUP (ssGBLUP) models for the genetic evaluation of Holstein (HO) and Jersey (JE) cattle with their crossbreds (CROSS). The dataset included 23,736,975 records on 8,560,986 cows. Genotypic data on 181,379 JE, 1,905,292 HO, and 53,799 CROSS animals were used for the evaluation. The genetic evaluation included 5 production traits, namely, milk yield (MY), protein yield (PY), fat yield (FY), SCS, and daughter pregnancy rate (DPR), which were analyzed using a 5-trait repeatability model using ssGBLUP with or without MF. Three different MF scenarios were tested: 4 MF (based on breed), 24 MF (based on the combination of breed, sex, and year of birth), and 32 MF (similar to 24 MF but with CROSS as a separate genetic group). The 3 MF scenarios were compared with a conventional ssGBLUP model that did not include metafounders (NO_MF). Forward-in-time validation was carried out to evaluate predictability, inflation, and stability. For purebred Holstein and Jersey cows, the truncated dataset included phenotypes through December 2018, whereas for crossbreds, the cutoff was December 2015; the complete dataset extended through December 2022. Validation targeted genotyped cows lacking records in their respective truncated dataset but with at least one record in the complete dataset, yielding 96,295 Holsteins, 26,436 Jerseys, and 5,099 crossbreds for analysis. Results showed that including MF affected prediction metrics differently depending on the trait, breed, and MF configuration. While certain MF classifications (e.g., 4 MF) reduce bias and improve predictability in crossbreds for some traits, others show minimal effects, particularly in purebred Holsteins. For low h² traits (SCS, DPR), MF scenarios provided better predictive ability in CROSS animals. In contrast, for high h² traits (MY, PY, FY), stability tended to decrease in MF models, suggesting possible overfitting due to added model complexity. Overall, MF offers a promising strategy to address pedigree gaps in multibreed evaluations, but its application should be carefully tailored to trait architecture and population composition to avoid overfitting and ensure accurate genetic predictions.