A review of deep learning applications for genomic selection

Area of Science:

• Genomics
• Machine Learning
• Plant and Animal Breeding

Background:

• Conventional genomic prediction methods, such as the additive genetic effect model, rely on mixed model equations.
• Deep learning (DL) methods offer a nonparametric approach with flexibility for complex data associations.

Purpose of the Study:

• To review deep learning applications in genomic selection (GS).
• To provide guidance on effective DL use in plant breeding.
• To compare DL with traditional genomic prediction approaches.

Main Methods:

• Review of current literature on DL in GS.
• Discussion of DL fundamentals and requirements.
• Comparative analysis of DL versus traditional methods.

Main Results:

• DL algorithms efficiently capture nonlinear genetic patterns.
• DL can integrate diverse data sources for GS.
• No clear superiority in prediction power over conventional models found in current literature.

Conclusions:

• Sufficiently large and high-quality training data are crucial for DL.
• DL shows potential for improving prediction accuracy in large-scale breeding data.
• DL algorithms are effective at integrating multi-source data for GS.