Robustly enhancing crop genomic prediction accuracy through ensemble learning and iterative optimization

Area of Science:

• Agricultural Science
• Genomics
• Bioinformatics

Background:

• Food security is threatened by climate change and population growth, necessitating faster breeding of superior crop varieties.
• Genomic prediction using genome-wide markers is crucial for intelligent crop breeding but faces challenges in cross-crop and cross-trait accuracy.
• Existing genomic prediction models often lack stable and accurate performance, hindering their widespread application.

Purpose of the Study:

• To develop a robust and accurate genomic prediction method for genotype-to-phenotype prediction.
• To improve the stability and accuracy of genomic prediction across different crops and traits.
• To provide an advanced tool for accelerating crop breeding programs.

Main Methods:

• Proposed GEG2P, a genetic algorithm-based ensemble learning method for genotype-to-phenotype prediction.
• Integrated 20 base learners with dynamic combination selection and weight optimization via a genetic algorithm.
• Utilized SHAP (SHapley Additive exPlanations) to analyze SNP contributions to phenotype prediction.

Main Results:

• GEG2P improved prediction accuracy by an average of 4.02% compared to the best single base learners.
• Demonstrated improved performance across multiple crops including maize, wheat, rice, chickpea, and soybean.
• Identified functional complementarity of SNPs with large effects captured by different base learners.

Conclusions:

• GEG2P offers a robust and accurate solution for genomic prediction in crop breeding.
• The ensemble learning approach enhances prediction stability and accuracy across diverse genetic backgrounds and traits.
• This method has the potential to significantly accelerate the development of improved crop varieties for global food security.