A Novel Strategy for Discovering the Best Genotype from a Large Number of Candidates with Genomic Prediction

Exploitation of useful genetic variations from germplasm collections preserved in gene banks is an important challenge for enhancing the potential of breeding. For a complex trait, genomic selection (GS) is expected to be a useful tool for searching the variations. When a part of accessions are phenotyped and genotyped to be used for building a prediction model, genotypic values of the remainders can be predicted from their marker genotypes based on the model. This enables to discover useful accessions from a large number of germplasm with a small cost of phenotyping. In this study, we will propose a novel selection criterion to discover the best accession from a large number of candidates efficiently and rapidly based on the genomic prediction. Usually, accessions with superior predicted values are selected from the top in the context of GS. However, when the model is not accurate, this greedy strategy fails to discover the best accessions. We introduced the expected improvement (EI) of genotypic values as a novel selection criterion. EI depends not only on the predicted genotypic values but also on their uncertainties, i.e. variances of the genomic prediction. A simulation study to evaluate the performance of EI criterion suggested that EI strategy can discover the best accession more rapidly than ordinal GS strategy.