A Cluster-Based Technique to Group Random Effects from Empirical BLUPs

Abstract

Prediction of random effects arises in many applications, including plant breeding. Plant breeders strive to select genotypes according to predictions of unobserved genetic merits derived fromphenotypic data. There are several approaches to predict random effects, but a popular one is to use empirical BLUPs (EBLUP) of genetic effects to rank genotypes. The problem is how best to use EBLUP values to select a set of truly superior genotypes with high probability. In this paper, we propose an algorithm based on hierarchical cluster analysis to group genotypes according to their EBLUPs. Clustering algorithms produce no analytical expression to solve the problem of detecting the number of underlying groups. Via simulation, we obtain decision curves to group genotypes from an EBLUP-based dendrogram. The new algorithm has good operating characteristics and can be used to select genotypes with high genetic merit.