BWGS: A R package for genomic selection and its application to a wheat breeding programme.

Area of Science:

• Plant breeding and genetics
• Bioinformatics and computational biology
• Quantitative genetics

Background:

• Genomic selection (GS) is crucial for accelerating crop improvement.
• Accurate prediction of breeding values requires robust computational tools.
• The BreedWheat project aimed to enhance wheat breeding efficiency through genomics.

Purpose of the Study:

• To develop an integrated R library, BWGS (BreedWheat Genomic selection), for easy computation of Genomic Estimates of Breeding Values (GEBV).
• To provide a flexible tool for cross-validation, GEBV prediction, and evaluation of different genomic prediction methods.
• To assess the impact of missing data, marker selection, and training population size on prediction accuracy in wheat.

Main Methods:

• Development of the BWGS R library utilizing existing R packages.
• Implementation of functions for replicated random cross-validation and GEBV prediction.
• Testing BWGS with a wheat population (760 lines, 47,839 SNPs), including missing data imputation and 15 prediction models.

Main Results:

• BWGS efficiently computed GEBV on a standard desktop computer, yielding comparable results to previous studies.
• Training population size and marker numbers significantly influenced predictive ability for traits with moderate heritability.
• Imputation handled up to 40% randomly distributed missing data without performance loss; up to 80% was acceptable with Expectation-Maximization.
• Marker selection improved predictive ability when applied to the whole population but not when limited to the training set, indicating overfitting.
• Minimal differences were observed among the 15 prediction models, with non-parametric methods showing slight advantages.

Conclusions:

• BWGS is a simple, powerful, and efficient R library for genomic selection in applied breeding programs.
• The study confirms theoretical predictions regarding factors influencing predictive ability in genomic selection.
• Imputation strategies and appropriate marker selection are vital for maximizing predictive accuracy in genomic selection.
• The high correlation among GEBV from different models suggests robustness and applicability in breeding.