Estimation of breeding values from large-sized routine carcass data in Japanese Black cattle using Bayesian analysis

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Bioinformatics

Background:

• Rapidly increasing volumes of genetic data necessitate efficient analysis methods.
• Current genetic evaluation methods like restricted maximum likelihood (REML) and empirical best linear unbiased prediction (EBLUP) have high memory requirements.
• A need exists for alternative approaches with smaller memory footprints for large-scale genetic evaluations.

Purpose of the Study:

• To apply and validate a Bayesian analysis using Gibbs sampling (GS) for genetic evaluation using large routine carcass field data.
• To compare the accuracy and memory efficiency of GS with the conventional REML-EBLUP approach.
• To confirm the suitability of GS for large datasets in genetic improvement programs.

Main Methods:

• A Bayesian analysis employing Gibbs sampling (GS) was implemented on a large dataset of Japanese Black routine carcass field data.
• Posterior means for breeding values were calculated after discarding initial samples and using a subset of the generated samples.
• The GS method was compared against the established REML-EBLUP method for accuracy and memory usage.

Main Results:

• The Gibbs sampling (GS) approach required only one-sixth of the memory space compared to REML-EBLUP.
• Moment and rank correlations between breeding values estimated by GS and REML-EBLUP were very close to one.
• Linear regression analysis showed coefficients close to one and intercepts close to zero, indicating strong agreement between the two methods.

Conclusions:

• The Bayesian analysis using Gibbs sampling (GS) is a valid and memory-efficient alternative for genetic evaluation using large routine carcass datasets.
• GS provides accurate breeding value estimations comparable to REML-EBLUP.
• This approach offers practical advantages for managing and analyzing extensive genetic datasets in livestock breeding programs.