Mitigating the impact of selective phenotyping in training populations on the prediction ability by multi-trait pedigree and genomic selection models

Training populations for pedigree and genomic prediction in plant breeding programmes are largely updated with superior genotypes from multi-environment trials, where they are tested with the goal of variety development. Such a selective phenotyping has, however, a negative impact on prediction abilities, especially when only a subset of breeding lines can be tested, as for laborious and costly to phenotype traits. This study focused, thus, on investigating the impact of selective phenotyping in the training population of an applied wheat breeding programme, and assessing the potential to mitigate this impact by pedigree and genomic multi-trait prediction models as well as bi-directionally selected training populations for several baking quality parameters. Combining both pedigree and genomic information in multi-trait prediction models with pre-existing phenotypic information for protein content and sedimentation value compensated for the observed loss in prediction ability, while entering few inferior breeding lines into the training population further mitigated the impact of selective phenotyping and even led to a slight increase of prediction ability in comparison to a randomly chosen training population.