Reliability of genomic predictions across multiple populations

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Genomic Prediction

Background:

• Genomic prediction utilizes dense SNP genotypes for predicting phenotypes, disease risk, and genetic merit in livestock and plants.
• Prediction reliability, the squared correlation with true genetic merit, indicates explained genetic variance and heavily relies on phenotype numbers.
• Combining datasets from multiple populations can increase reliability, especially with scarce phenotypes, but may decrease it if marker effects differ significantly.

Purpose of the Study:

• To assess the impact of combining multiple populations on genomic prediction reliability.
• To investigate how population divergence (T = 6, 30, 300 generations) and training set composition affect prediction accuracy.
• To determine the role of marker density in maintaining prediction reliability across divergent populations.

Main Methods:

• Simulated two cattle populations (A and B) with varying divergence times (T).
• Constructed training sets by combining individuals from population A (1000) with varying numbers from population B (0-1000).
• Varied marker density and trait heritability to evaluate their effects on genomic prediction reliability.

Main Results:

• Adding population B individuals increased reliability in population A by up to 0.12 (high marker density, T=6).
• Conversely, reliability in population A decreased by up to 0.07 (low marker density, T=300).
• Reliability in population B, initially lower (up to 0.77) without population A data, improved significantly with combined data when marker density was high enough for linkage disequilibrium to persist.

Conclusions:

• Combining phenotypes from all populations generally yields the most accurate genomic predictions.
• For highly diverged populations, a higher marker density is crucial for reliable genomic predictions.
• The effectiveness of combining populations depends on balancing the benefits of increased sample size against potential decreases in reliability due to differing marker effects.