Multi-trait and multi-environment genomic prediction for flowering traits in maize: a deep learning approach.

Area of Science:

• Plant breeding and genetics
• Genomics
• Computational biology

Background:

• Maize (Zea mays L.) is a vital global food security crop.
• Identifying genomic regions for agronomic traits enhances breeding efficiency.
• Flowering traits (ASI, FF, MF) are crucial for maize development.

Purpose of the Study:

• Compare deep learning and Bayesian models for genomic prediction accuracy.
• Evaluate multi-trait and multi-environment approaches.
• Identify genomic regions associated with flowering time in tropical maize.

Main Methods:

• Utilized a tropical maize panel (258 lines) with ~290,000 SNPs.
• Applied deep learning and Bayesian models (MCMCglmm, BGGE, BMTME).
• Evaluated models across multiple traits and environments.

Main Results:

• Multi-trait models increased prediction accuracy by 14.4% over single-trait, single-environment.
• Multi-environment analysis improved accuracy by 6.4% over multi-trait.
• Deep learning models consistently outperformed Bayesian models.

Conclusions:

• Deep learning models offer superior prediction accuracy for genomic selection in tropical maize.
• These models enhance the selection of superior genotypes for flowering traits.
• Findings support deep learning's efficacy in maize breeding programs.