Bias in genomic predictions for populations under selection

Area of Science:

• Animal breeding and genetics
• Quantitative genetics
• Genomic selection

Background:

• Genome-wide marker-assisted selection (GWMAS) is crucial for livestock and plant breeding.
• Accurate and unbiased genomic predictions are essential for successful GWMAS.

Purpose of the Study:

• To investigate the impact of selection intensity on the accuracy and bias of genomic predictions.
• To compare the performance of single-step and multiple-step genomic prediction methods.

Main Methods:

• Simulated two animal populations under varying heritabilities and selection strengths (weak and strong).
• Utilized best-linear unbiased prediction (BLUP) with both multiple-step (pseudodata) and single-step (combined relationship matrices) approaches.
• Evaluated prediction bias and accuracy under different selection scenarios.

Main Results:

• The multiple-step method produced biased predictions.
• The single-step method showed reduced bias and higher accuracy but was less accurate under strong selection.
• Adjusting genomic relationships in the single-step method with a constant derived analytically eliminated bias and maximized accuracy.

Conclusions:

• The single-step method offers superior accuracy and reduced bias in genomic predictions compared to the multiple-step method.
• Accounting for selection bias by adjusting genomic relationships is critical for accurate genomic evaluations, particularly in strongly selected populations.