Maximizing efficiency in sunflower breeding through historical data optimization.

Area of Science:

• Plant breeding
• Genomics
• Statistical genetics

Background:

• Genomic selection (GS) is widely used in plant breeding due to lower genotyping costs and increased computational power.
• Extensive historical genotypic and phenotypic data necessitate strategies for optimal utilization in GS models.
• Across-year prediction accuracy is crucial for developing robust breeding programs.

Purpose of the Study:

• To investigate optimal data subset selection for calibrating GS models for across-year predictions.
• To develop and evaluate methods for optimizing training set size and genetic composition.
• To enhance predictive ability of GS models using historical data in sunflower breeding.

Main Methods:

• A multi-objective optimization approach was used to select ideal training set years.
• The Min_GRM method was developed to optimize training set size, reducing dimensionality.
• The Tails_GEGVs algorithm was employed to optimize genetic composition and leverage heterogeneity.

Main Results:

• The Min_GRM method reduced dimensionality by 20% with minimal loss in predictive ability.
• Tails_GEGVs outperformed using all data, utilizing only 60% for grain yield prediction.
• Maximizing genetic diversity in the training set ensured consistent predictive ability across genotypic values.

Conclusions:

• Optimal utilization of historical data through strategic subset selection can improve GS model performance.
• The developed optimization methods, Min_GRM and Tails_GEGVs, offer efficient ways to enhance predictive ability.
• This research provides valuable insights for maximizing the effectiveness of GS in plant breeding programs.