Normalized cumulative gain as an alternative evaluation measure for genomic selection models

Area of Science:

• Genomics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic selection (GS) uses statistical and machine learning models to predict phenotypes from genomic data.
• Comparing GS model performance is crucial as no single method excels across all datasets.
• Traditional metrics like Pearson's correlation and MSE assess overall fit but not selection effectiveness for breeding.

Purpose of the Study:

• To introduce an alternative evaluation metric, Normalized Cumulative Gain (NCG), for genomic prediction.
• To assess the effectiveness of GS models in selecting top-performing individuals for breeding programs.

Main Methods:

• Developed and applied the Normalized Cumulative Gain (NCG) metric.
• Compared nine common genomic prediction methods across four animal and plant datasets.
• Analyzed model performance across all possible selection thresholds.

Main Results:

• NCG directly quantifies the phenotypic gain from selected individuals, offering an intuitive measure of selection efficiency.
• Performance comparisons revealed differences in method effectiveness under varying selection intensities.
• Comprehensive analysis across all thresholds provides more insight than single-threshold evaluations.

Conclusions:

• NCG provides a more relevant evaluation for breeding selection than traditional regression metrics.
• The choice of selection intensity significantly impacts method performance, guiding optimal model selection.
• The R package for NCG calculation is available for broader application in genomic selection research.