How Genetic Variance and Number of Genotypes and Markers Influence Estimates of Genomic Prediction Accuracy in Plant Breeding

Genomic prediction is revolutionizing plant and animal breeding, but its accuracy is affected by multiple factors. Here, we simulate 24 scenarios, each with 1000 datasets, to evaluate how varying the genetic variance (small, large), number of genotypes (180, 360, 540, and 698), and markers (2912, 5823, and 11,646) affects the relative performance of seven competing methods for accuracy estimation in genomic prediction in plant breeding programs. Each method was used to estimate predictive accuracy. The estimates were then compared between methods and, for each method, with the simulated true accuracy for each scenario as the gold standard. The genetic variance and the number of genotypes and markers strongly and jointly influenced estimation accuracy. Accuracy was highest when the genetic variance was large and the numbers of genotypes (n = 698) and markers (n = 11,646) were highest. A recently proposed method (Method 5) and a method commonly used in animal breeding (Method 7) produced the most globally accurate, precise, and stable estimates of accuracy. Among the methods that use cross-validation (Methods 1-4 and 6), Method 4 gave the most stable estimates of accuracy. Reducing genetic variance whilst increasing the numbers of genotypes and markers considerably prolonged the computing time for all methods. Thus, for quantitative traits with sizable genetic variances, using about 700 genotypes and 12,000 markers and using Method 5 or 7 should result in accurate genomic prediction in plant breeding.