Robust estimation of heritability and predictive accuracy in plant breeding: evaluation using simulation and...

Area of Science:

• Plant breeding and genetics
• Statistical modeling in agriculture
• Genomic selection and prediction

Background:

• Genomic prediction (GP) is crucial for identifying superior genotypes in plant breeding.
• Accurate estimation of predictive accuracy is vital for effective GP.
• Classical regression models in plant breeding are sensitive to outliers, leading to biased estimates of heritability and predictive accuracy.

Purpose of the Study:

• To evaluate the performance of robust statistical methods compared to classical approaches for estimating heritability and predictive accuracy in genomic prediction.
• To assess the impact of data contamination with outliers on the reliability of these estimation methods.

Main Methods:

• Comparison of robust and classical regression approaches for analyzing phenotypic data.
• Simulation studies under various outlier contamination scenarios (random and block).
• Application to commercial maize and rye breeding datasets.

Main Results:

• The robust approach generally matched or exceeded the classical approach in accuracy.
• Robust methods consistently outperformed classical methods under random data contamination.
• Empirical dataset analyses confirmed the stability and reliability of the robust approach with outliers and missing data.

Conclusions:

• The proposed robust approach effectively enhances heritability and genomic prediction accuracy by mitigating outlier effects.
• Plant breeders are encouraged to integrate robust methods alongside classical approaches for improved accuracy.
• Increasing the number of replicates to three or more can further boost the accuracy of the robust approach.