AMMI Bayesian models to study stability and adaptability in maize

Abstract

The identification of genotypes presenting wide adaptability and stability is pivotal in breeding programs. To identify such genotypes, it is necessary to use sophisticated analytical tools to establish the genotypes × environments interaction (GEI) pattern across multi-environment trials and select for genotypic stability and adaptability. The aim of the present study was to estimate GEI using Bayesian analysis of Additive Main Effects and Multiplicative Interaction (AMMI) models for both balanced and unbalanced data sets and estimate the predictive ability of model. Two studies were assessed to showcase this approach; in the first, 10 commercial maize (Zea mays) single-cross hybrids and 45 double-cross hybrids were evaluated at 15 different locations. In the second study, 28 hybrids were evaluated in 35 different environments distributed over two different harvest seasons (first and second harvests) with unbalanced data sets within and between harvests. The Bayesian analysis of the AMMI models was robust in dealing with the unbalanced data. This approach is promising for the identification of interaction patterns and the estimation of GEI. The genotypes and environments could be grouped according to their interaction patterns even using the unbalanced data sets, showing that Bayesian analysis of AMMI models could be applied effectively for multi-environment trials. The prediction for missing hybrids was satisfactory in a simulated unbalanced design and captured the GEI and patterns in the data. This allowed the direct comparison of genotypes from the first and second harvests and the estimation of selection gain.