Accuracies of genomic prediction for twenty economically important traits in Chinese Simmental beef cattle

Bo Zhu, Peng Guo, Zhen Wang, W. Zhang, Yan Chen, Lei Zhang, Huijiang Gao, Xue Gao, Lingyang Xu, J. LiAnimal 2019 · DOI

TL;DR

This study evaluated five genomic prediction methods for 20 economically important traits in Chinese Simmental beef cattle. Bayesian regression methods, particularly BayesR and BayesCπ, showed slightly superior predictive accuracies compared to the standard GBLUP method, suggesting their potential for future genomic selection in this breed.

What Problem Does This Paper Address?

Genomic selection, which uses an animal's DNA information to predict its breeding value, has revolutionized animal breeding by accelerating genetic gain. However, the accuracy of these predictions is crucial for effective selection. While various statistical methods exist for genomic prediction, their performance can vary depending on the traits, breed, and available data. This paper addresses the need to evaluate and compare different genomic prediction approaches for a comprehensive set of economically important traits (growth, carcass, and meat quality) in the Chinese Simmental beef cattle population, a breed significant in China. Understanding which methods perform best in this specific context is vital for optimizing breeding programs and improving beef production efficiency.

Methods

The researchers employed five statistical methods to estimate genomic breeding values (GEBVs):

  • GBLUP (Genomic Best Linear Unbiased Prediction): A widely used method that assumes all SNPs contribute to the genetic variance.
  • Bayesian Regression Methods:
    • BayesA: Assumes each SNP has its own variance.
  • BayesB: Assumes a mixture of distributions, with some SNPs having zero effect and others having non-zero effects.
  • BayesCπ: Similar to BayesB but with a common variance for SNPs with non-zero effects.
  • BayesR: An extension of BayesCπ, where SNP effects are sampled from a mixture of normal distributions with fixed variances.

These methods were applied using a five-fold cross-validation approach. This involved repeatedly splitting the data into training (reference) and testing (validation) sets to assess how well the models predicted phenotypes for unseen individuals. Predictive ability was evaluated using correlations between predicted and observed phenotypes, as well as regression coefficients to assess prediction bias.

Datasets

The study utilized phenotypic and genotypic data from Chinese Simmental beef cattle.

  • Phenotypic Data: Collected from animals born between 2008 and 2013, encompassing 20 economically important traits including growth (e.g., average daily gain, live weight), carcass traits (e.g., carcass weight, lean meat percentage, striploin weight), and meat quality traits (e.g., pH, shear force).
  • Genotypic Data: Obtained using the Illumina BovineHD SNP array, which provides a high density of genetic markers. Missing genotypes were imputed.

This dataset represents a specific population of Chinese Simmental cattle, and the findings are specific to this genetic background and the traits measured.

Key Insights and Ideas

  • Bayesian Methods Show Promise: The study highlights that Bayesian regression methods, particularly BayesR and BayesCπ, generally offer slightly better predictive accuracies than the traditional GBLUP for many traits. This suggests that models that can account for a variable number of SNPs with significant effects might be more adept at capturing the genetic architecture of these traits.
  • Trait-Specific Performance: Predictive accuracies varied significantly across the 20 traits. Growth and carcass traits tended to have moderate to high accuracies, while meat quality traits showed lower accuracies. This is likely due to differences in heritability and the genetic complexity of these traits.
  • Heritability Matters: As expected, traits with higher heritability generally exhibited higher predictive accuracies. The study visually demonstrated a positive correlation between predictive ability and heritability across the different methods.
  • Prediction Scale and Bias: The study observed that predictions from most methods tended to be inflated (overestimated) for many traits, though some traits showed deflation. BayesR and BayesCπ showed a tendency towards less inflation compared to GBLUP and BayesA, suggesting they might provide more reliable scaled predictions.

Key Results

  • Predictive Accuracies: The predictive accuracies (measured as the correlation between GEBV and corrected phenotype, adjusted for heritability) ranged from 0.159 (lean meat percentage with BayesCπ) to 0.518 (striploin weight with BayesR).
  • Average Accuracies: The average predictive accuracies across all 20 traits were: GBLUP (0.361), BayesA (0.361), BayesB (0.367), BayesCπ (0.367), and BayesR (0.378). BayesR showed the highest average accuracy.
  • Regression Coefficients (Scale): The average regression coefficients across traits were: GBLUP (0.89), BayesA (0.86), BayesB (0.89), BayesCπ (0.94), and BayesR (0.95). Higher values closer to 1 indicate less bias in the scale of predictions. BayesR and BayesCπ demonstrated less bias.
  • Mean Squared Error (MSE): While GBLUP often had lower MSE for some traits, Bayesian methods (BayesB, BayesCπ, BayesR) showed lower MSE for several other traits, indicating a better overall fit for those specific traits.

Limitations and Future Work

  • Reference Population Size: The authors acknowledge that increasing the size of the reference population is crucial for improving prediction accuracies.
  • Inflation of Predictions: While BayesR and BayesCπ showed less inflation than other methods, the issue of overestimation in GEBVs persists for many traits. Further research is needed to address this bias.
  • Trait Complexity: The lower accuracies for meat quality traits suggest that these traits might be genetically more complex or influenced by more subtle genetic effects, requiring further investigation.
  • Future Application: The study concludes that Bayesian regression approaches, particularly BayesR and BayesCπ, are feasible for future genomic selection in Chinese Simmental beef cattle, especially with larger reference populations. They suggest that further improvements in population size and bias reduction are necessary for optimal implementation in commercial breeding.

Summary updated

Abstract

Genomic prediction has been widely utilized to estimate genomic breeding values (GEBVs) in farm animals. In this study, we conducted genomic prediction for 20 economically important traits including growth, carcass and meat quality traits in Chinese Simmental beef cattle. Five approaches (GBLUP, BayesA, BayesB, BayesCπ and BayesR) were used to estimate the genomic breeding values. The predictive accuracies ranged from 0.159 (lean meat percentage estimated by BayesCπ) to 0.518 (striploin weight estimated by BayesR). Moreover, we found that the average predictive accuracies across 20 traits were 0.361, 0.361, 0.367, 0.367 and 0.378, and the averaged regression coefficients were 0.89, 0.86, 0.89, 0.94 and 0.95 for GBLUP, BayesA, BayesB, BayesCπ and BayesR respectively. The genomic prediction accuracies were mostly moderate and high for growth and carcass traits, whereas meat quality traits showed relatively low accuracies. We concluded that Bayesian regression approaches, especially for BayesR and BayesCπ, were slightly superior to GBLUP for most traits. Increasing with the sizes of reference population, these two approaches are feasible for future application of genomic selection in Chinese beef cattle.
BibTeX
@inproceedings{zhu2019accuracies,
  author = {Bo Zhu and Peng Guo and Zhen Wang and W. Zhang and Yan Chen and Lei Zhang and Huijiang Gao and Zhen Wang and Xue Gao and Lingyang Xu and J. Li},
  title = {{Accuracies of genomic prediction for twenty economically important traits in Chinese Simmental beef cattle}},
  booktitle = {Animal},
  year = {2019},
  doi = {10.1111/age.12853},
}
Figure