Methods to approximate reliabilities in single-step genomic evaluation.

Area of Science:

• Animal Breeding and Genetics
• Quantitative Genetics
• Bioinformatics

Background:

• Calculating prediction reliability in single-step genomic best linear unbiased prediction (ssGBLUP) using matrix inversion is computationally intensive for large datasets.
• Existing methods for reliability calculation in genomic evaluations face scalability challenges.

Purpose of the Study:

• To develop and evaluate computationally feasible approximation methods for ssGBLUP prediction reliability.
• To assess the accuracy of these approximations for large genotyped populations.

Main Methods:

• Developed two approximation methods for reliability based on decomposing reliability functions into contributions from records, pedigrees, and genotypes.
• Tested approximation methods using a simulated dataset, comparing results to ssGBLUP.
• Included empirical correction for inflation in approximated reliabilities.

Main Results:

• The first approximation method showed high correlation (0.98) with ssGBLUP reliabilities, improving to 0.99 after correction.
• The second approximation method achieved a correlation of 0.72, increasing to 0.89 after correction.
• Both methods proved accurate and computationally feasible for populations up to 100,000 genotyped animals.

Conclusions:

• Approximation methods for ssGBLUP reliability are accurate and computationally efficient for large-scale genomic evaluations.
• Quality control of single nucleotide polymorphism (SNP) data and proper scaling of the genomic relationship matrix are crucial for approximation accuracy.
• These methods enable reliable genomic predictions in large livestock and other breeding populations.