AMMI-Bayesian models and use of credible regions in the study of combining ability in maize

Abstract

The development of lines with high performance and stability for synthesis of superior hybrids is the most expensive and time-consuming phase in maize hybrid breeding. Several times, due to available resources, only a part of possible hybrid combinations is tested. Therefore, the breeder needs methods that allow the evaluation of genotypes untested in the field. This work was carried out with the objective of proposing a prediction model of general and specific (SCA) combining ability, and interactions with environments, associated with the use of credible regions in biplots obtained through Additive Main Effects and Multiplicative Interaction Bayesian model. Two analyses were done, in which the first one was conducted with simulated data, and the second one with real data. Credible ellipses were constructed in biplot in order to evaluate the stability of interaction effects for GCA and SCA. For the analysis of simulated data, the predictions obtained had high correlation with the real values. For the effects of GCA and SCA, the predictions kept the standard of signals and rank. The model was efficient to provide credible intervals which covered the simulated values. For the analysis of real data, estimates of GCA and SCA for all genotypes evaluated do not differ from zero. The biplots for GCA × E and SCA × E interactions allowed evaluate the genotype stability in a more accurate way and the uncertainty about interaction estimates. The model is shown as a promising tool for helping the breeder to select and recommend genotypes.