Sparse Phenotyping and Haplotype-Based Models for Genomic Prediction in Rice

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Multi-environment genomic selection (MEGS) is crucial for developing resilient crop varieties.
• Robust training datasets with multi-environment phenotypic data are essential for effective MEGS.
• Genomic prediction enhanced sparse phenotyping offers cost savings in multi-environment trials (MET).

Purpose of the Study:

• To investigate the effectiveness of multi-environment training sets with varying phenotyping intensities.
• To evaluate different haplotype-based genomic prediction models using LD-derived haplotype blocks.
• To enhance multi-environment genomic selection in rice breeding.

Main Methods:

• Utilized three rice populations with diverse sizes and compositions.
• Employed LD-derived haplotype blocks for haplotype-based genomic prediction.
• Assessed prediction accuracy for days to heading (DTH) and plant height (PH) under varying phenotyping intensities.

Main Results:

• Phenotyping only 30% of records in the training set yielded comparable prediction accuracy to high-intensity phenotyping.
• Local epistatic effects appear significant for DTH.
• Smaller haplotype blocks (2-3 SNPs) maintained predictive ability in large populations.
• Modeling inter-environment covariances improved genomic prediction accuracy.

Conclusions:

• Sparse phenotyping is efficient for building multi-environment training sets in rice.
• Haplotype-based models, particularly with optimized block sizes, are effective for MEGS.
• Accounting for environmental correlations enhances prediction accuracy in rice breeding programs.