Genome position specific priors for genomic prediction.

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic prediction accuracy relies heavily on reference population size.
• Pooling data from different breeds can improve accuracy but is less effective for distantly related breeds.
• Existing methods struggle with genetic variations across populations.

Purpose of the Study:

• To introduce BayesRS, a novel method for sharing genomic information across populations.
• To improve genomic prediction accuracy, particularly for small or distantly related populations.
• To assess the effectiveness of BayesRS in dairy cattle breeds.

Main Methods:

• BayesRS derives SNP effect prior proportions from one population for use in another.
• The model accounts for reversed SNP allele phases and different causative mutations affecting the same gene.
• Tested on Australian Jersey, Australian Holstein, and Nordic Holstein cattle for protein, fat, and milk yield traits using 777K SNPs.

Main Results:

• BayesRS increased Jersey population prediction accuracy by up to 3.5% compared to models without location-specific priors.
• Improvements were significant for Australian Holstein (1-2% accuracy increase) for protein and fat yield.
• Accuracy gains were generally lower than traditional data pooling, except for fat yield in Jerseys.

Conclusions:

• BayesRS can be advantageous over data pooling for distantly related populations for certain traits.
• The method provides increased accuracy compared to single-population analysis without added computational burden.
• BayesRS offers a flexible framework for incorporating biological information into genomic predictions.