Fine tuning genomic evaluations in dairy cattle through SNP pre-selection with the Elastic-Net algorithm

Area of Science:

• Animal Genetics
• Statistical Genomics
• Bioinformatics

Background:

• Genomic selection aims to estimate single-nucleotide polymorphism (SNP) effects for breeding value prediction.
• The challenge of "p >> n" (more SNPs than individuals) can lead to poor SNP effect estimation in methods like genomic best linear unbiased prediction (gBLUP).

Purpose of the Study:

• To evaluate the performance of the Elastic-Net (EN) algorithm for genomic selection in dairy cattle.
• To investigate the impact of SNP pre-selection based on quantitative trait locus (QTL) detection on prediction accuracy and computational efficiency.

Main Methods:

• Comparison of EN, gBLUP, and pedigree-based BLUP using data from three French dairy cattle breeds.
• Application of SNP pre-selection using QTL detection to both EN and gBLUP methods.
• Validation of direct genomic values (DGV) against observed daughter yield deviations.

Main Results:

• The EN algorithm demonstrated encouraging performance compared to gBLUP and pedigree-based BLUP.
• SNP pre-selection did not significantly alter prediction accuracy (correlation between DGV and observed values).
• SNP pre-selection substantially reduced the number of SNPs used in the EN prediction equation, decreasing computational load.

Conclusions:

• Elastic-Net is a promising approach for genomic selection, effectively handling the "p >> n" problem.
• SNP pre-selection, particularly with QTL detection, offers significant computational advantages for large-scale genomic evaluations without compromising accuracy.
• These findings support the feasibility of efficient national genetic evaluations with increasing numbers of animals and SNPs.