Comparing algorithms to approximate accuracies for single-step genomic best linear unbiased predictor.

Area of Science:

• Animal breeding and genetics
• Quantitative genetics
• Statistical genomics

Background:

• Calculating exact accuracy of estimated breeding values requires inverting the left-hand side (LHS) of mixed model equations (MME).
• Inverting the LHS is computationally infeasible for large datasets, particularly with genomic information.
• Approximation algorithms are necessary to estimate accuracies in large-scale genetic evaluations.

Purpose of the Study:

• To compare two algorithms (Algorithm 1 and Algorithm 2) for approximating accuracies within the BLUPF90 software.
• To validate these approximated accuracies against exact accuracies derived from MME inversion.
• To assess the impact of including genotyped animals (with and without phenotypes) on accuracy estimations.

Main Methods:

• Compared two approximation algorithms: Algorithm 1 (using genomic relationship matrix diagonal) and Algorithm 2 (combining accuracies with/without genomic data).
• Validated against exact accuracies from MME inversion on a subset of data using single-trait models.
• Utilized extensive datasets from the American Angus Association, including pedigree and genotype information for millions of animals across growth, carcass, and marbling traits.

Main Results:

• Algorithm 2 demonstrated higher correlations (0.98-0.99) with exact accuracies compared to Algorithm 1 (0.87-0.90) for genotyped animals.
• Algorithm 2 showed better performance with lower mean square errors for most traits and regression slopes closer to 1 (0.82-0.87) versus Algorithm 1 (0.98-1.10).
• Approximated accuracies from Algorithm 2 more closely mirrored exact accuracies when genotyped animals were included in the analysis.

Conclusions:

• Algorithm 2 is more precise and reliable for approximating accuracies in genetic evaluations, particularly with large genomic datasets.
• The inclusion of genotyped animals significantly benefits accuracy estimation, with Algorithm 2 effectively capturing these improvements.
• Algorithm 2 is recommended for practical genetic evaluations due to its superior accuracy and computational feasibility.