G2P Provides an Integrative Environment for Multi-model genomic selection analysis to improve genotype-to-phenotype...

Area of Science:

• Genomics
• Computational Biology
• Plant Breeding

Background:

• Genotype-to-phenotype (G2P) prediction is vital for genomic selection (GS)-assisted breeding.
• Existing GS models vary in prediction precision across species and traits, necessitating model evaluation and selection.

Purpose of the Study:

• To introduce the G2P container, an integrated environment for evaluating and selecting optimal GS models.
• To enhance G2P prediction accuracy through auto-ensemble algorithms and optimized training set design.

Main Methods:

• Developed the G2P container for the Singularity platform, including 16 state-of-the-art GS models and 13 evaluation metrics.
• Implemented parallel processing on high-performance computing clusters for comprehensive model evaluation.
• Utilized auto-ensemble algorithms for automatic model selection and integration of prediction results.
• Incorporated genetic diversity analysis for optimizing training set composition.

Main Results:

• The G2P container provides unbiased evaluation of 16 GS models.
• Auto-ensemble algorithms effectively select the most precise models and integrate predictions, improving G2P accuracy.
• Optimized training sets, based on genetic diversity, achieved prediction precision comparable to full sets with fewer samples.
• Reduced phenotyping costs by utilizing optimized training sets.

Conclusions:

• The G2P container offers a comprehensive solution for evaluating and improving GS models.
• Its auto-ensemble and training set optimization features enhance prediction accuracy and practical utility in breeding programs.
• The tool facilitates cost-effective phenotyping and accelerates genomic selection applications.