Training population design and resource allocation for genomic selection in plant breeding

Wide-scale adoption of genomic selection in cultivar development programs will dramatically alter the role of phenotyping from a direct source of information for selection to an indirect source of information through the training of genomic prediction models. While some features of phenotyping and field trials in plant breeding programs such as measurement of relevant phenotypes in relevant environments will remain the same, some features of plant breeding field trials will most certainly change. Potential changes include which types of individuals are phenotyped in field trials and how field plot resources are allocated to genotypes in order to maximize the amount and quality of information used for developing genomic prediction models. We provide a brief and intuitive review of the current literature on these two topics. By far, the most important consideration in training population design is the genetic relationship with the target population, making the definition of the target population the first and most important step in genomic selection model development. Several algorithms for training population design that show promise have been published and should be considered by practitioners. When the goal of phenotyping is to train genomic prediction models, plots should be allocated to maximize population size, although much more flexibility exists for genomic selection model training compared to QTL mapping and marker-assisted selection applications. Trials across environments can be connected through markers, making it possible to further maximize population size and allowing extensive capturing of genotype x environment interaction effects in early, preliminary yield trial stages.