Across population genomic prediction scenarios in which Bayesian variable selection outperforms GBLUP

Area of Science:

• Animal genetics
• Genomic prediction
• Statistical genetics

Background:

• Genomic prediction accuracy is limited in small populations.
• Across-population prediction offers potential but currently yields disappointing results.
• Bayesian variable selection (BVS) models show promise over GBLUP for low quantitative trait loci (QTL) numbers within populations.

Purpose of the Study:

• To identify scenarios where BVS models outperform GBLUP for across-population genomic prediction.
• To evaluate the impact of the number of QTL and genetic correlation on prediction accuracy.

Main Methods:

• Utilized high-density genotype data from Holstein Friesian, Groningen White Headed, and Meuse-Rhine-Yssel cows.
• Simulated phenotypes based on varying numbers of QTL (3000, 300, 30, 3) and genetic correlations (1.0, 0.8, 0.4).
• Compared prediction accuracies of BVS models and GBLUP.

Main Results:

• BVS model accuracy increased as the number of QTL decreased, a trend more pronounced in across-population prediction.
• BVS models outperformed GBLUP when the number of QTL was small, with the advantage diminishing as QTL number increased.
• The accuracy of BVS and GBLUP became comparable when the number of QTL equaled the effective number of independent chromosome segments (Me).

Conclusions:

• BVS models surpass GBLUP when the number of underlying QTL is less than Me.
• Me is larger across populations than within populations, increasing the likelihood of BVS superiority.
• BVS models offer a viable strategy to enhance across-population genomic prediction accuracy.