Genetic diversity among pepper and chili genotypes by Kohonen’s Self-Organizing Maps
The study of genetic diversity of a population is one of the pillars to make selection in a breeding program successful. There are several techniques capable of estimating genetic divergence. Among them, those based on multivariate statistics deserve attention. Recently, methodologies based on Artificial Neural Networks (ANN) have been used to study the genetic diversity of a population. One of the strategies within ANN is Kohonen’s self-organizing maps (SOMs), which allows the organization of genetic diversity. We estimated and organized the genetic divergence of pepper and chili genotypes Capsicum annum) for selection of diallel crosses. The experiment was conducted in a greenhouse in a completely randomized design with four replications. Nine commercially important genotypes of C. annum were evaluated, based on seven quantitative characters of the fruit. Univariate analysis was performed by analysis of variance and cluster mean. Multivariate [unweighted pair group method with arithmetic means (UPGMA) and Tocher)] and machine-learning techniques [SOMs] were employed. The genotypes showed high genetic variability for all traits. The traits total mass of raw fruit and fresh fruit length contributed the most to genetic diversity. UPGMA and Tocher classified the genotypes into two and four clusters, respectively. Through SOMs, it was observed that the neighborhood pattern between chili and pepper was obtained in only seven of the 12 neurons previously established. Overall, the use of SOM allowed the organization of genetic diversity among the genotypes. Specifically, SOM did not recommend crossing genotypes from the Cascadura Ikeda chili x Giant Ruby chili, Spicy for Pot pepper x Yellow Jamaica pepper, and Volcano pepper x Peter Pepper for variability exploration.