Predicting genetic variance in bi-parental breeding populations is more accurate when explicitly modeling the segregation of informative genomewide markers

Robust predictions of genetic variances for important traits would facilitate greater genetic gains in plant breeding. Previous attempts to predict the genetic variance ((Formula presented.)) of traits in bi-parental breeding populations were inconsistent and context specific. The weakness of methods that consider the phenotypic distance, genetic distance, and relationship-based distance of pairs of parents, which we collectively term historical methods, stems from the fact that they do not explicitly model the segregation of the underlying genetic effects for a trait within a population. To address this issue, we propose the use of three modern methods made possible by the commonplace use of genomewide molecular marker data and genomic selection in modern breeding programs. These modern methods utilize both phenotypic and genotypic records to, in varying degrees, explicitly model the segregation of informative genomewide markers to predict (Formula presented.) in bi-parental breeding populations. In this study, we evaluate the accuracy of historical and modern methods to predict (Formula presented.) using 40 field-tested bi-parental barley breeding populations evaluated during 2003–2010 for Fusarium head blight severity. In general, the modern methods predicted the field-based estimates of (Formula presented.) more accurately than the historical methods. Specifically, the modern method that most explicitly models the segregation of informative genomewide markers, called ‘PopVar,’ was the most accurate (Formula presented.) prediction method.