Classifying Mexican Maize Accessions Using Hierarchical and Density Search Methods

Abstract

Cluster analysis is commonly used for studying genetic diversity. Problems with hierarchical cluster analysis include how to combine different types of variables (discrete and continuous), choosing distance measurements, applying an appropriate clustering strategy, designating the optimal number of clusters, and identifying variables with significant discriminatory power. Hierarchical clustering methods are only descriptive and do not represent probabilities for classifying individuals into groups. The objectives of this study were to: (i) examine the performance of different cluster strategies based on several criteria, (ii) propose a classification method for germplasm accessions with statistical properties, and (iii) examine how the results of the proposed classification classification method can be applied to form core subsets. Morphologic and agronomic attributes collected for 115 Mexican maize (Zea mays L.) accessions, grouped in five races, from the Latin America Maize Project (LAMP) were subjected to the hierarchical cluster algorithms UPGMA (arithmetic mean method), Centroid, Median, and the Ward method. Two other techniques were studied, Density and the Normix (Nor) density search methods, which were both restricted continuous variables. The Nor method was applied to groups formed “a priori” by means of the hierarchical methods UPGMA, Centroid, Median, and Ward and resulted in subgroups denoted as NorU, NorC, NorM, and NorW, respectively. The NorW method formed five well defined groups of accessions and was an appropriate strategy for grouping accessions into relatively homogeneous groups. Strategies UPGMA, NorU, Centroid, NorC, Median, NorM, and Density were not very effective for classifying accessions into homogeneous groups. Different subsets can be formed based on the characteristics of the five homogeneous groups formed by NorW.