Predictor bias in genomic and phenomic selection.

Area of Science:

• Agricultural Science
• Plant Breeding
• Spectroscopy

Background:

• Genomic prediction is crucial for crop breeding progress.
• Phenomic prediction, using technologies like hyperspectral imaging, offers an alternative to predict breeding values without genetic markers.
• Near-infrared spectroscopy (NIRS) has a history of predicting compositional parameters and, more recently, grain yield.

Purpose of the Study:

• To compare the predictive ability of genomic prediction versus phenomic prediction using hyperspectral measurements of wheat grains.
• To evaluate phenomic prediction for various traits, including grain yield.
• To identify biases and limitations in phenomic prediction for wheat breeding.

Main Methods:

• Utilized hyperspectral measurements (NIRS) from wheat grains as phenomic predictors.
• Compared phenomic prediction models with traditional genomic prediction models.
• Assessed prediction accuracy for multiple traits, focusing on grain yield and protein content.

Main Results:

• Phenomic prediction outperformed genomic prediction for certain traits.
• Phenomic predictions for grain yield were found to be inflated and biased by grain protein content.
• Unbiased phenomic prediction abilities were considerably lower than previously reported.

Conclusions:

• NIRS of wheat grains is a biased predictor for grain yield, primarily reflecting protein content.
• Phenomic prediction shows potential but requires methods to ensure unbiasedness and retain population parameters.
• Future research should focus on developing unbiased phenomic prediction strategies for effective application in crop breeding programs.