From linear models to deep learning: statistical advances in genomic selection for animal breeding

Area of Science:

• Animal Breeding and Genetics
• Bioinformatics
• Computational Biology

Background:

• Genomic selection (GS) accelerates genetic gain in animal breeding using genome-wide markers.
• Increasing data dimensionality and complexity challenge traditional linear models in GS.
• Advances in genotyping and sequencing technologies provide high-density genomic data.

Purpose of the Study:

• To provide a comprehensive overview of Genomic Selection (GS) methodologies.
• To compare traditional linear and Bayesian models with modern machine learning (ML) and deep learning (DL) approaches.
• To offer practical guidance for optimizing genomic evaluation strategies in the era of big data.

Main Methods:

• Review of statistical foundations of linear mixed and Bayesian models.
• Survey of machine learning (ML) and deep learning (DL) algorithms for genomic data.
• Comparative analysis of computational efficiency and predictive accuracy across methods.

Main Results:

• Traditional models assume additive effects and may oversimplify complex genomic relationships.
• ML and DL algorithms effectively capture nonlinear relationships, addressing limitations of classical methods.
• The study compares the performance of diverse GS approaches for practical application.

Conclusions:

• ML and DL offer powerful alternatives to traditional models for genomic evaluation, especially with complex, high-dimensional data.
• Understanding the assumptions, advantages, and limitations of each method is crucial for optimizing breeding strategies.
• This review guides the selection of appropriate GS methodologies for maximizing genetic gain in modern animal breeding programs.