Feature engineering of environmental covariates improves plant genomic-enabled prediction

Area of Science:

• Agricultural Science
• Genetics
• Bioinformatics

Background:

• Genomic selection (GS) is crucial for modern breeding but requires high prediction accuracy for practical application.
• Factors influencing GS prediction performance necessitate strategies for improvement in breeding programs.
• Incorporating environmental covariates into GS models does not always enhance prediction performance.

Purpose of the Study:

• To explore feature engineering of environmental covariates to improve genomic prediction models.
• To assess the impact of feature engineering on prediction error and accuracy in genomic prediction.

Main Methods:

• Feature engineering was applied to environmental covariates.
• Genomic prediction models were utilized with and without engineered environmental covariates.
• Prediction performance was evaluated across multiple datasets.

Main Results:

• Feature engineering reduced prediction error by 761.625% compared to using environmental covariates without feature engineering.
• Significant gains in prediction accuracy were observed in specific datasets.
• The effectiveness of feature engineering varied across different datasets.

Conclusions:

• Feature engineering shows promise for enhancing genomic prediction accuracy.
• Further research is required to establish robust feature engineering strategies for environmental covariates.
• Optimizing the incorporation of environmental data can advance genomic selection in breeding programs.