RESUMO
This study aimed to evaluate the clustering pattern consistency of soybean (Glycine max) lines, using seven different clustering methods. Our aim was to evaluate the best method for the identification of promising genotypes to obtain segregating populations. We used 51 generations F5 and F6 soybean lines originating from different hybridizations and backcrosses obtained from the soybean breeding program of Universidade Federal de Uberlândia in addition to three controls (Emgopa 302, BRSGO Luziânia, and MG/BR46 Conquista). We evaluated the following agronomic traits: number of days to flowering, number of days to maturity, height of the plant at maturity, insertion height of the first pod, grain yield, and weight of 100 seeds. The data was submitted to analyses of variance followed by average Euclidean distance matrix estimation used as measure of dissimilarity. Subsequently, clusters were formed using the Tocher method and dendrograms were constructed using the hierarchical methods simple connection (nearest neighbor), complete connection (most distant neighbor), Ward, median, average within cluster connection. The nearest neighbor method presented the largest number of genotypes in group I and showed the greatest similarity with the Tocher optimization method. The joint use of these two methodologies allows for differentiation of the most genetically distant genotypes that may constitute the optimal parents in a breeding program.