Genomic prediction using subsampling.

Area of Science:

• Genomics
• Animal Breeding
• Plant Breeding

Background:

• Genome-wide assisted selection is crucial for genetic improvement in plants and animals.
• Whole-genome regression models in a Bayesian framework are standard for genomic prediction.
• Fitting these models with large datasets presents significant computational challenges.

Purpose of the Study:

• To introduce and evaluate the subsampling bootstrap Markov chain method for genomic prediction.
• To assess the impact of subsampling on both prediction accuracy and computational efficiency.

Main Methods:

• The study proposes a subsampling bootstrap Markov chain approach for fitting whole-genome regression models.
• This method involves subsampling observations within each round of a Markov Chain Monte Carlo simulation.
• The impact of subsampling on prediction and computational parameters was evaluated across various datasets.

Main Results:

• An optimal subsampling proportion of approximately 50% with replacement and 33% without replacement was identified.
• Subsampling reduced model fitting time by approximately 50%.
• Losses in predictive properties due to subsampling were negligible, typically less than 1%, with occasional slight improvements observed.

Conclusions:

• The subsampling bootstrap Markov chain algorithm effectively reduces the computational burden of model fitting in genomic prediction.
• Combining subsampling with Gibbs sampling forms an effective ensemble algorithm.
• This method shows potential for enhancing prediction properties while significantly improving computational efficiency.