Sensitivity of genomic selection to using different prior distributions

Area of Science:

• Animal genetics
• Quantitative genetics
• Bioinformatics

Background:

• Genomic selection (GS) leverages dense genetic markers, like single nucleotide polymorphism (SNP) data, to predict genomic estimated breeding values (GEBV).
• Various Bayesian models exist for deriving GS prediction equations, primarily differing in their specification of prior distributions.

Purpose of the Study:

• To evaluate the impact of different prior distributions within Bayesian models on the accuracy of GEBV predictions.
• To compare the performance of four distinct Bayesian approaches for genomic selection using a simulated dataset.

Main Methods:

• Analysis of a simulated dataset from the 13th QTL-MAS workshop.
• Application of four Bayesian methods to predict GEBV for individuals lacking phenotypic data.
• Systematic variation of prior distributions to assess their influence on prediction accuracy.

Main Results:

• All tested Bayesian methods generated GEBV that exhibited high correlation with true breeding values.
• The accuracy of predicted GEBV was largely insensitive to the specific choice of prior distributions used in the models.
• Consistent performance across different methods was observed, aligning with findings from real-world data.

Conclusions:

• The choice of prior distributions in Bayesian models has a limited effect on GEBV prediction accuracy for the analyzed QTL-MAS dataset.
• The robustness of these models suggests reliable performance across various prior specifications in genomic selection.
• Findings indicate a degree of uniformity in the performance of different Bayesian methods for genomic prediction.