Multibreed genomic prediction using multitrait genomic residual maximum likelihood and multitask Bayesian variable...

Area of Science:

• Animal Genetics and Breeding
• Quantitative Genetics
• Genomic Prediction

Background:

• Genomic prediction models are applicable across different breeds, but multi-breed information has shown limited benefits.
• Investigating advanced models to leverage multi-breed data for improved genomic prediction accuracy is crucial.

Purpose of the Study:

• To evaluate a multitask Bayesian stochastic search variable selection (BSSVS) model for multi-breed genomic prediction.
• To compare the performance of multitask BSSVS against single-trait BSSVS, GREML, and pooled data approaches.

Main Methods:

• Implemented a multitask Bayesian stochastic search variable selection (BSSVS) model to accumulate evidence of quantitative trait loci (QTL) across breeds.
• Compared BSSVS and genomic residual maximum likelihood (GREML) models using single-trait, multi-trait, and pooled data from Holstein and Jersey bulls.
• Genotyped 474,773 single nucleotide polymorphisms (SNPs) and used phenotypes for milk, fat, and protein yields in a training dataset of 6,278 Holstein and 722 Jersey bulls.

Main Results:

• Single-trait BSSVS consistently outperformed GREML. Multitask BSSVS did not outperform single-trait BSSVS using pooled data.
• Pooling Holstein and Jersey data significantly increased prediction accuracy for milk and protein yields in Jerseys.
• The Bayesian model identified QTL in Holsteins, outperforming GREML, while multitask models did not surpass simple pooling strategies.

Conclusions:

• A simple pooling strategy assuming equal trait effects across breeds yielded superior genomic prediction accuracy compared to complex multitask models.
• High genetic correlations for milk and protein yields, potentially due to breed admixture, favor pooling strategies.
• The effectiveness of pooling suggests that shared moderate- to large-effect QTL across breeds are critical for multi-breed genomic prediction.