Balancing Sensitivity and Specificity Enhances Top and Bottom Ranking in Genomic Prediction of Cultivars

Area of Science:

• Agricultural Science
• Genetics
• Biotechnology

Background:

• Genomic selection (GS) is revolutionizing plant and animal breeding by enabling predictive selection of desirable traits.
• Accurate identification of top-performing lines is critical for the practical application of GS, but remains challenging.
• Existing GS methods vary in efficiency, necessitating a comparison of approaches for precise top-performer selection.

Purpose of the Study:

• To benchmark and compare the performance of five different genomic selection models for identifying top (or bottom) performing lines.
• To evaluate the effectiveness of a streamlined tuning approach for methods requiring rigorous tuning.
• To identify the most accurate and efficient GS methodology for practical breeding applications.

Main Methods:

• Five models were evaluated: Bayesian Best Linear Unbiased Predictor (GBLUP, RC), GBLUP with a threshold (R), Regression Optimum (RO), Threshold Bayesian Probit Binary (TGBLUP, B), and TGBLUP with an optimal threshold (BO).
• A benchmark comparison was conducted using five real datasets.
• A streamlined tuning approach was proposed and applied for methods requiring tuning.

Main Results:

• The Regression Optimum (RO) method demonstrated superior performance across all five datasets, outperforming R, B, RC, and BO in F1 score by significant margins.
• RO also showed marked improvements in Kappa coefficient and Sensitivity compared to other models.
• The BO model was identified as the second-best performing method, with both BO and RO optimizing thresholds to balance Sensitivity and Specificity.

Conclusions:

• The Regression Optimum (RO) and Optimal Threshold Bayesian Probit Binary (BO) methods are superior for selecting top (or bottom) performing lines in genomic selection.
• These advanced methods, particularly RO, offer significant improvements in accuracy and efficiency over traditional approaches.
• Breeders are encouraged to adopt the BO and RO methods to enhance genetic gain in breeding programs.