Animal welfare

Sensitivity of genomic heritability and predictive ability to SNP-panel reduction in dairy cattle personality traits

---

Abstract

Phenotyping personality traits in dairy cattle is inherently time-consuming, labor-intensive and poorly standardized across studies. Consequently, available datasets for genetic analyses remain comparatively small, which limits statistical power and prevents a comprehensive understanding of the genetic architecture of such traits. At the same time, genomic prediction requires sufficiently dense SNP information to capture additive genetic variance. Yet, it remains unclear to what extent SNP density can be reduced, and which reduction strategy is most robust without compromising the estimation of genomic based heritability or prediction accuracy for personality traits. This study assessed the sensitivity of genomic based heritability (h²) estimates and genomic predictive ability to alternative SNP-panel reduction strategies for four personality traits (explorative, bold, sociable, trusting). These traits were derived from a principal component analysis of behavioral responses measured in a novel-object test and a forced-human-approach test. Data were obtained from 306 Holstein cows genotyped with a 50k SNP array. Four SNP-panels were generated: a full post-QC panel (40,165 SNPs), an LD-pruned panel (8,324 SNPs), an LD-clustered panel (9,346 SNPs), and a random-thinned panel (9,000 SNPs). Genomic relationship matrices were computed using the VanRaden method. Single-trait GREML analyses were performed in the R package sommer, including fixed effects for lactation number, days-in-milk class and test batch. Prediction accuracies were obtained via repeated 5-fold cross-validation with 10 repetitions using the R package rrBLUP. Heritability estimates from the full panel ranged from 0.07 (explorative) to 0.41 (trusting). The LD-pruned and random-thinned panels yielded highly similar results (0.066-0.475 and 0.073-0.404, respectively). The LD-clustered panel produced comparable estimates for boldness, sociability and trusting (e.g., trusting = 0.446), whereas the estimate for the explorative trait markedly decreased (0.038). Predictive abilities showed analogous patterns: full-panel accuracies ranged from 0.03 to 0.20, and both the LD-pruned and random panels showed minimal reductions. In contrast, the LD-clustered panel preserved accuracy for three traits but yielded a correlation near zero for the explorative trait. These findings demonstrate that SNP-panel reduction to approximately 9k markers can be implemented with negligible loss of precision in heritability estimation or genomic prediction for most personality traits. Considering that behavioral phenotyping typically produces limited sample sizes, placing genomic analyses in a p > > n framework, the results emphasize the relevance of marker-reduction strategies to maintain estimator stability and predictive ability.

Keywords: 2026

How to Cite:

Seidel, A., Hasenpusch, P., Wilder, T. & Thaller, G., (2026) “Sensitivity of genomic heritability and predictive ability to SNP-panel reduction in dairy cattle personality traits”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2287018. doi: https://doi.org/10.31274/wcgalp.24197

Rights: 1