GWAS & Selection signatures

Comparison of SNP marker effect estimates of single-step and multi-step genomic models using millions of genotyped animals of German Holstein

---

Abstract

A single-step SNP BLUP model has been used for official genomic evaluation of German Holstein since April 2025, replacing the previous multi-step SNP BLUP model. Earlier comparison of the two genomic models was focused on genomic breeding values of diverse groups of animals. The objective of this study was to compare estimates of SNP marker effects of the two genomic models for a total of 42 traits from 6 trait groups. Phenotypic, genotypic and pedigree data were obtained from the official evaluation in December 2024, at which the multi-step genomic evaluation was conducted for the last time. Genotype records of 1,702,070 Holstein animals were used in both single-step and multi-step evaluations. Formilk prodcution traits, the total number of cows with test-day records amounted to 14,209,131 and the number of integrated bulls having foreign daughters in MACE evaluation was 143,135. The pedigree files for the single-step evaluation contained 22,931,615 animals for each of four traits and 42,562,772 for calving traits. In comparison to the single-step model, a mixed reference population of cows and bulls was used in the routine SNP effect estimation under the previous multi-step SNP BLUP model for German Holstein. In the official multi-step genomic evaluation in December 2024, the total number of reference cows and bulls reached 609,598 for milk yield. Because the single-step model for German Holstein assumed 30% residual polygenic variance across all the evaluated traits but the residual polygenic variance differed between the traits under the multi-step model, we set here the proportion of residual polygenic variance of the multi-step model equal to 30% for all the evaluated traits as for the single-step model. SNP effects of the multi-step model were re-estimated for all the selected traits assuming 30% residual polygenic variance. Pearson correlation of the SNP effect estimates between the two SNP BLUP models ranges from 0.67 for direct genetic effect of stillbirth to 0.87 for foot angle and body condition score with a mean of 0.81 for all 42 traits. The single-trait model of SNP effect estimation in the multi-step evaluation appears to contribute mostly to the correlations between the two genomic models that are lower than unity. Variance of the SNP effect estimates is slightly greater for the multi-step model than single-step model for most of the traits. Based on a truncated validation data set for 23 conformation traits, we demonstrated that the single-step model gave a higher accuracy (average correlation 0.94 vs 0.86) and lower prediction bias (average regression slope 1.02 vs 0.93) of the SNP effect estimates. This study on comparing SNP effect estimates provides valuable insights into the superiority of the single-step over multi-step models in terms of prediction accuracy and bias of genomic estimated breeding values for animals.

Keywords: 2026

How to Cite:

Liu, Z., Alkhoder, H., Polman, L., Heise, J. & Reents, R., (2026) “Comparison of SNP marker effect estimates of single-step and multi-step genomic models using millions of genotyped animals of German Holstein”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2281879. doi: https://doi.org/10.31274/wcgalp.23461

Rights: 1