Do feature selection methods for selecting environmental covariables enhance genomic prediction accuracy?

Area of Science:

• Agricultural Science
• Genetics
• Bioinformatics

Background:

• Genomic selection (GS) is revolutionizing breeding but faces challenges with complex traits and multi-environmental data.
• Integrating environmental covariates into GS models is crucial for improving prediction accuracy.

Purpose of the Study:

• To investigate the integration of environmental information with genotypic data in genomic selection.
• To evaluate the effectiveness of feature selection methods (Pearson's correlation and Boruta) for this integration.

Main Methods:

• Applied two feature selection techniques to identify relevant environmental covariates.
• Utilized a leave-one-environment-out cross-validation strategy.
• Assessed prediction accuracy using Normalized Root Mean Squared Error (NRMSE) and Pearson's correlation.

Main Results:

• Simple inclusion of environmental covariates had variable effects on prediction accuracy.
• Optimal incorporation via feature selection significantly improved prediction accuracy in 4 out of 6 datasets (14.25%–218.71% NRMSE improvement).
• No significant gain in Pearson's correlation was observed; feature selection was ineffective when covariates were unrelated to the trait.

Conclusions:

• Feature selection is a valuable strategy for optimizing the integration of environmental covariates in genomic selection.
• This approach empirically enhances prediction power for complex traits under multi-environmental conditions.
• The effectiveness depends on the relevance of environmental covariates to the target trait.