Some objective functions and ideas to optimize experimental designs in artificial selection programs

Abstract

Experimental designs in artificial selection programs must balance limited resources with the need for accurate genetic evaluation across diverse environments. We propose a decision-point framework that decomposes the overall design process into sequential allocation steps, each optimized using objective functions based on representativeness. Using stochastic simulations, we evaluated how these objective functions influence accuracy and genetic gain across multiple design decisions. Key results emerged. First, the selection of a sufficiently large and representative set of environments had the greatest impact on across target population of environments (TPEs) accuracy. When the TPE was complex, sparse-testing strategies consistently outperformed fully balanced designs, even when total plot numbers were held constant. Second, environment selection using a modified optimal-contribution function (−q′Dq) outperformed random sampling, k-means, and hierarchical clustering, yielding up to a 20% gain in accuracy and reduced uncertainty. Third, once environments were fixed, the allocation of entries across environments had a moderate but meaningful effect, especially when genetic correlations varied widely across the TPE. Fourth, within-environment design decisions strongly influenced accuracy: optimal field size depended on the level of spatial heterogeneity, while replication had relatively limited impact when genomic relationship matrices were used. Finally, optimized spatial allocation improved accuracy under strong spatial noise, though at increased computational cost. Overall, this framework confirms prior findings while offering a practical and computationally efficient approach for optimizing phenotypic evaluations. By tailoring objective functions to specific decision points, breeding programs can improve accuracy, reduce risk, and make better use of limited resources.