A simple method to separate base population and segregation effects in genomic relationship matrices

Area of Science:

• Animal Breeding and Genetics
• Quantitative Genetics
• Genomics

Background:

• Genomic selection and genomic breeding values (GBV) are crucial in livestock and plant breeding.
• Investigating population subdivision via the genomic relationship matrix (G) is common, but its impact on GBV estimation using genomic best linear unbiased prediction (GBLUP) is understudied.

Purpose of the Study:

• To develop and validate a method to decompose the genomic relationship matrix (G) to understand the influence of population structure on GBV estimation.
• To compare different GBLUP models incorporating base population structure.

Main Methods:

• Decomposed G into components representing allele frequency differences (G A(*)) and within-group relationships (G S).
• Utilized Fst statistics to assess genetic distances and tested three GBLUP models (M0, M1, M2) in a forward prediction scenario with cattle data.
• Analyzed base group contributions (Q) and estimated effects and prediction errors.

Main Results:

• The decomposition revealed significant differences in base group effects within breeds.
• Forward prediction showed minimal differences in reliability between models, but M1 (using only G S) had the lowest predictive power.
• The best model (M0 or M2) depended on breed, trait, and validation group composition.

Conclusions:

• Standard GBLUP using the G-matrix is equivalent to a model (M0) that treats base groups as correlated random effects.
• The proposed decomposition aids in dissecting the contributions of base population structure and genetic divergence to genomic relationships and breeding values.