The Impact of Sparse Testing design on Prediction in Wheat: An Empirical Study Using T3/Wheat data

Sparse testing is a multi-environment trial design in which different subsets of all accessions are tested in each environment. Particularly at early stages of breeding programs, when many accessions need to be evaluated, sparse testing can be an efficient way to allocate resources to maximize genetic gain under genotype x environment interactions. Appropriate sparse testing designs can increase the prediction accuracy of unobserved accession x environment combinations. This study was done to evaluate the prediction accuracy of four sparse testing designs: pure sparse (Mpure), overlap sparse (Mover), sparse with common accessions (Mcomm), and overlap sparse with common accessions (Movco) using five groups of trials sourced from the Wheat instance of The Triticeae Toolbox database. The MegaLMM package was used to predict the performance of untested accessions within a known environment using the factor analytic model. Mover, Mcomm, and Movco gave better prediction accuracy than Mpure. Across all tested scenarios, the use of 90 accessions per trial as the training set increased the prediction accuracy as compared to 60 accessions per trial. Given these results, we recommend the use of sparse testing with some level of overlapping accessions across trials to evaluate breeding lines at an early stage of the breeding program and to impute missing accession x environment combinations.