Assessing Genomic Selection Prediction Accuracy in a Dynamic Barley Breeding Population

Area of Science:

• Plant breeding
• Quantitative genetics
• Genomic selection

Background:

• Genomic selection (GS) prediction accuracy is typically assessed via simulation and cross-validation.
• Real-world validation using progeny performance in dynamic plant breeding programs remains less explored.
• Understanding prediction accuracy over time is crucial for optimizing breeding strategies.

Purpose of the Study:

• To evaluate prediction model performance using progeny data in a barley breeding population.
• To investigate the impact of training population composition on prediction accuracy over time.
• To identify factors influencing prediction accuracy in a dynamic breeding context.

Main Methods:

• Evaluated multiple prediction models, including random regression best linear unbiased prediction (RR-BLUP), in a 647-line barley population.
• Utilized 1536 single nucleotide polymorphism (SNP) markers across four traits with varying genetic architectures.
• Trained models on a parent set and five consecutive progeny sets spanning a 5-year period.

Main Results:

• Minimal differences in prediction accuracy were observed among models, with RR-BLUP performing consistently well.
• Using only the parent set for training was largely sufficient for predicting progeny performance.
• Larger training populations combining parent and progeny data yielded negligible accuracy improvements.
• Prediction accuracy varied widely (0.03-0.99) across traits and progeny sets.
• Marker allele frequency, population structure, and linkage disequilibrium influenced accuracy.
• Reduced prediction accuracy for the DON trait correlated with marker fixation over time.
• Higher trait heritability (H²) and simpler genetic architecture in training populations enhanced prediction accuracy.

Conclusions:

• The parent set alone provides a robust basis for genomic prediction in this barley breeding program.
• Expanding training populations with subsequent progeny offers limited benefits for prediction accuracy.
• Trait heritability and genetic architecture are key determinants of successful genomic prediction.