Computing strategies for multi-population genomic evaluation.

Area of Science:

• Animal breeding and genetics
• Quantitative genetics
• Genomic selection

Background:

• Genomic prediction across multiple breeds aims to improve accuracy by utilizing diverse population data.
• Incorporating non-additive effects like dominance and genotype-by-environment interactions increases model complexity.
• Current methods, such as single nucleotide polymorphism best linear unbiased prediction (SNP-BLUP) and genomic best linear unbiased prediction (GBLUP), face computational challenges with complex models.

Purpose of the Study:

• To evaluate computational efficiency of different genomic prediction models for multi-breed evaluations.
• To propose and assess an alternative approach using generalized least squares (GLS) and selection index (SI) theory.

Main Methods:

• Comparison of SNP-BLUP and GBLUP formulations for multi-breed genomic prediction.
• Development and application of a generalized least squares (GLS) followed by selection index (SI) backsolving method.
• Evaluation of computational demands (number of operations and memory) for each method.

Main Results:

• SNP-BLUP directly provides marker effects but involves a large number of equations with complex block structures.
• GBLUP is general but can be memory-intensive due to unused equations and dense/sparse blocks.
• The proposed GLS+SI method is the least computationally demanding in operations and memory, utilizing single dense blocks, but requires specialized programming.

Conclusions:

• Standard Henderson's mixed model equations become less practical for multi-breed genomic predictions with increasing inclusion of non-additive effects.
• The GLS+SI approach offers a more computationally efficient alternative for complex genomic prediction scenarios.
• This method is particularly relevant as multi-breed evaluations and non-additive genetic effects become more prevalent.