Optimizing the performance of large genomic evaluations through data truncation in Angus cattle

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic breeding values (GEBV) are crucial for genetic selection in livestock populations.
• Large genomic datasets increase computational costs in single-step GBLUP models.
• Optimizing data usage is essential for balancing accuracy and computational efficiency.

Purpose of the Study:

• To evaluate the impact of phenotypic and genotypic data truncation on computing time and prediction accuracy.
• To determine if reduced datasets can yield comparable results to full datasets in large genomic evaluations.
• To assess the efficiency of indirect predictions for excluded genotyped animals.

Main Methods:

• Applied data truncation strategies to phenotypic (birth year) and genotypic (animal records/progeny) data in the American Angus growth model.
• Analyzed traits including birth weight (BW), weaning weight (WW), and post-weaning gain (PWG).
• Validated GEBV using LR and predictive ability methods, comparing full datasets (P-all, G-all) with truncated datasets (e.g., P-2005, G-info).
• Calculated indirect predictions (IP) for genotyped animals excluded from the main evaluation.

Main Results:

• Moderate data truncation (P-2005/G-info) reduced computing time by 66% with no significant loss in prediction accuracy.
• LR prediction accuracy ranged from 0.62-0.63 (BW), 0.74-0.77 (WW), and 0.72-0.74 (PWG) across scenarios.
• Correlations between GEBV and indirect predictions (IP) were high (≥ 0.99), indicating similar means and accuracies.
• Truncation showed minimal impact on prediction accuracy and dispersion, especially with robust data structures and medium-to-high heritability traits.

Conclusions:

• Phenotypic and genotypic data truncation is a viable strategy to reduce computational burden in large-scale genomic evaluations.
• Moderate truncation effectively balances prediction accuracy and computational cost, enhancing the efficiency of genetic selection.
• Indirect predictions offer a fast and reliable method for estimating genetic merit in non-informative animals.