Improving the efficiency of genomic selection.

Area of Science:

• Quantitative Genetics
• Animal Breeding
• Plant Breeding
• Genomic Selection

Background:

• Genomic selection (GS) is crucial for predicting phenotypes from genome-wide markers in breeding programs.
• Increasing the efficiency of phenotypic prediction is essential for cost-effective and accurate breeding.
• Current GS methods can be computationally intensive and require extensive genotyping data.

Purpose of the Study:

• To investigate two novel approaches for enhancing the efficiency of phenotypic prediction in genomic selection.
• To evaluate the impact of feature selection using Markov blankets on GS model performance and cost.
• To compare different kinship coefficient calculations for genomic best linear unbiased prediction (GBLUP), including a novel LD-adjusted measure.

Main Methods:

• Applied feature selection based on Markov blankets to identify informative genome-wide markers for GS models.
• Compared four GS models (partial least squares, ridge regression, LASSO, elastic net) with and without feature selection.
• Evaluated various kinship coefficients for GBLUP, including standard methods and a new linkage disequilibrium (LD)-adjusted kinship measure.
• Utilized three real-world datasets with continuous phenotypic traits from plant and animal genetics for performance assessment.

Main Results:

• Feature selection using Markov blankets resulted in simpler GS models with no loss, and potentially improved, predictive power.
• Elastic net combined with feature selection demonstrated strong performance across the evaluated datasets.
• Genomic best linear unbiased prediction (GBLUP) using the novel LD-adjusted kinship measure also performed exceptionally well.
• Both elastic net with feature selection and LD-adjusted GBLUP were identified as the top-performing methods in the study.

Conclusions:

• Feature selection via Markov blankets offers a theoretically sound and practical approach to simplify GS models and potentially reduce genotyping costs.
• The choice of kinship coefficients significantly impacts GBLUP performance, with LD-adjusted measures providing superior accuracy.
• The combination of elastic net with feature selection and LD-adjusted GBLUP represents a highly effective strategy for efficient and accurate genomic prediction in breeding.