Fast-forwarding plant breeding with deep learning-based genomic prediction.

Area of Science:

• Plant breeding and genetics
• Bioinformatics
• Machine learning in agriculture

Background:

• Deep learning-based genomic prediction (DL-based GP) offers advantages over traditional methods for complex multi-omics data in plant breeding.
• Current DL-based GP methods face challenges including data requirements, performance benchmarking, and environmental factor integration.

Purpose of the Study:

• To summarize key obstacles in developing DL-based GP models.
• To propose future research directions for advancing DL-based GP.

Main Methods:

• Review and synthesis of existing literature on DL-based GP in plant breeding.
• Identification of critical challenges and limitations in current methodologies.

Main Results:

• Key obstacles identified: need for large, high-quality datasets; inconsistent performance benchmarking; difficulties in integrating environmental factors.
• Proposed future directions: modular DL approaches, data augmentation techniques, and advanced attention mechanisms.

Conclusions:

• Addressing these obstacles is crucial for the effective development and widespread adoption of DL-based GP.
• Future research should focus on innovative solutions like modularity and enhanced data handling to improve predictive accuracy and utility in plant breeding programs.