# Genomic Prediction and Heritability Estimation for Daughter Pregnancy Rate in U.S. Holstein Cows Using SNP, Epistasis and Haplotype Effects

## TL;DR

This study explores how different genetic effects influence the prediction of pregnancy rates in U.S. Holstein cows, finding that additive and additive-by-additive interactions are most important.

## Contribution

The study introduces the use of additive × additive epistasis effects to improve genomic prediction accuracy for daughter pregnancy rate in Holstein cows.

## Key findings

- Additive × additive (A × A) epistasis effects had nonzero heritability and improved prediction accuracy by 4.88%.
- Inter-chromosome A × A effects were primarily responsible for prediction accuracy and phenotypic variance.
- Larger sample sizes significantly increased prediction accuracy, with peak accuracy in samples of 90,000 cows.

## Abstract

The contributions of additive, dominance, haplotype, and epistasis effects up to the third order to the accuracy of predicting daughter pregnancy rate (DPR) phenotypic values and to the phenotypic variance in U.S. Holstein cows were investigated using five samples with 25,827–133,934 cows and 74,855–75,209 SNPs. Heritability estimates showed that only additive × additive (A × A) epistasis effects had nonzero heritability and all other second- and third-order epistasis effects had zero heritability, and hence A × A was the only epistasis effects included in the prediction models. Based on the results of the largest sample with 133,934 cows, genomic heritability estimate was 0.044–0.054 for additive heritability, 0.005 for dominance heritability, 0.011–0.022 for haplotype heritability, and 0.052–0.062 for A × A heritability. The combination of additive (A) and A × A effects was the best prediction model based on the prediction accuracy. This best model improved the prediction accuracy over the A-only model by 4.88%, and had total heritability of 0.099 as the summation of the additive and A × A heritability estimates. Dominance and haplotype effects had minor contributions (0.97–2.44%) to prediction accuracy in models without A × A effects but had no contribution to prediction accuracy when A × A was in the prediction model. The partition of A × A effects into inter- and intra-chromosome A × A effects showed that inter-chromosome A × A were mainly responsible for the A × A contributions to prediction accuracy and phenotypic variance. Sample size had a major impact on prediction accuracy and the sample of 90,000 cows or 81,000 cows per training population had peak prediction accuracies that were 32.70–35.85% higher than in the sample with 25,827 cows. The largest sample with 133,934 cows had the smallest variations in prediction accuracy and slightly lower average prediction accuracy than in the sample with 90,000 cows.