BWGS

BWGS computes Genomic Estimated Breeding Values (GEBV) and implements genomic selection analyses for crop breeding, including wheat, using SNP marker data.

Key Features:

• Replicated Random Cross Validation: Performs replicated random cross-validation within training sets composed of both genotyped and phenotyped lines.
• GEBV Prediction: Predicts GEBVs for sets of lines that are genotyped but lack phenotype data.
• Missing Data Imputation: Provides missing-data imputation methods that retain predictive ability with up to 40% randomly distributed missing genotypes and up to 80% missingness when using the Expectation-Maximization method from the rrBLUP package.
• Marker and Training Set Selection: Enables selection of markers and training sets, including marker selection across the entire population to target markers associated with traits and reduce overfitting from sample-based selection.
• Diverse Genomic Prediction Methods: Implements 15 genomic prediction methods encompassing parametric and semi-parametric approaches.

Scientific Applications:

• Wheat genomic selection case study: Applied to adjusted yield data from historical trials with highly unbalanced designs involving 760 candidate lines genotyped at 47,839 robust SNPs.

Methodology:

Computational methods explicitly include replicated random cross-validation on combined genotyped/phenotyped training sets; GEBV prediction for genotyped-only lines; missing-data imputation including the Expectation-Maximization method from rrBLUP; marker selection across the entire population; implementation of 15 parametric and semi-parametric genomic prediction models; assessment of predictive ability relative to training population size and number of markers to capture QTL; and evaluation of non-parametric methods for non-additive effects.