Abordagem e bayesiana na avaliação genética de plantas perenes e modelos lineares generalizados aplicados na seleção de cultivares e no mapeamento de QTLs

Abstract

Bayesian procedures and Generalized Linear Models theory (GLM) have been indicated as inference methodologies appropriated for genetic analysis. These have been recently used, for instance: for diversity and phylogeny studies, and mapping of Quantitative Trait Loci (QTL). Bayesian inference is also considered as an alternative method of the classic Mixed Linear Models. However, the use of these procedures in the plant breeding programs has been relatively restricted. Thus, the analytical objective of the present study was to examine several experiments about plant breeding using such methodologies, and including variants of the GLM approach: Mixed-GLM and Generalized Estimating Equations (GEE). The study confirmed both the effectiveness and broad applicability of GLM, in experiments about plant breeding. In the current study, by using GLM, it was possible: 1. Mapping of QTLs controlling binary traits, which were measured repeatedly on the same subject (longitudinal analysis), 2. Mapping of QTLs controlling continuous trait, but biased from the normal distribution, and considering the principle of composite interval method, and 3. Evaluating field trials with olive cultivars, in situations where the response variables follow the Gama (fruit production) and Binomial (early fruit production and survival) distributions. It was confirmed that the information about agronomical data distribution should be stressed in the plant breeding programs, to improve the reliability of the genetic inference. The Bayesian approach was found to provide practical information useful for breeding purposes and for understanding how quantitative traits are controlled genetically in olive, acacia and eucalyptus. Binary traits, which were used to analyze plant survival, production precocity and early flowering, were also included in the breeding programs by using threshold models and the Monte Carlo Markov Chain (MCMC) variants. The ability of Bayesian inference to predict breeding values of clones, provenances, families and individual plants (animal model) was found to be very valuable for the genetic evaluation of plants