Multi-trait/environment sparse genomic prediction using the SFSI R-package.

Area of Science:

• Quantitative genetics
• Genomic prediction
• Statistical modeling

Background:

• Sparse selection indices (SSIs) predict genetic merit using high-dimensional phenotypes.
• Sparse genomic prediction (SGP) predicts genetic merit using subsets of training data.
• Existing methods do not fully integrate variable and data subset selection.

Purpose of the Study:

• Introduce a novel multi-trait/environment sparse genomic prediction (MT-SGP) framework.
• Combine the strengths of SSIs and SGP into a unified model.
• Provide an R-package for implementing SSIs, SGP, and MT-SGP.

Main Methods:

• Developed an MT-SGP framework integrating SSI and SGP principles.
• Utilized an R-package for solving SSI, SGP, and MT-SGP problems.
• Conducted extensive benchmarks using three diverse datasets (crops, traits, environments).

Main Results:

• MT-SGP demonstrated improved or comparable prediction accuracy to MT-GBLUP (up to 15% gain).
• Identified key factors influencing MT-SGP performance: sample size, genetic correlation, and heritability.
• The R-package provides practical tools for applying these sparse prediction methods.

Conclusions:

• MT-SGP offers a powerful approach for enhancing genetic merit prediction accuracy.
• The framework effectively borrows information from relevant traits and genetically similar individuals.
• MT-SGP provides a valuable alternative to traditional genomic prediction methods, especially under specific genetic architectures.