Development and application of statistical genetic methods to genomic prediction in Coffea canephora

• Ano de defesa: 2017
• Autor(a) principal: Luís Felipe Ventorim Ferrão
• Orientador(a): Antonio Augusto Franco Garcia
• Banca de defesa: Luciana Lasry Benchimol, Carlos Augusto Colombo, Gabriel Rodrigues Alves Margarido, Luiz Filipe Protasio Pereira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade de São Paulo
• Programa de Pós-Graduação: Agronomia (Genética e Melhoramento de Plantas)
• País: BR
• Link de acesso: https://doi.org/10.11606/T.11.2017.tde-17082017-143756

Genomic selection (GS) works by simultaneously selecting hundreds or thousands of markers covering the genome so that the majority of quantitative trait loci are in linkage disequilibrium (LD) with such markers. Thus, markers associated with QTLs, regardless of the significance of their effects, are used to explain the genetic variation of a trait. Simulation and empirical results have shown that genomic prediction presents sufficient accuracy to help success in breeding programs, in contrast to traditional phenotypic analysis. For this end, an important step addresses the use of statistical genetic models able to predict the phenotypic performance for important traits. Although some crops have benefited from this approach, studies in the genus Coffea are still in their infancy. Until now, there have been no studies of how predictive models work across populations and environments or, even, their performance for different complex traits. Therefore, the main objective of this research is investigating important aspects related to statistical modeling in order to enable a more comprehensive understanding of what makes a robust prediction model and, as consequence, apply it in practical breeding programs. Real data from two experimental populations of Coffea canephora, evaluated in two brazilian locations and SNPs identified by Genotyping-by-Sequencing (GBS) were considered to investigate the genotype-phenotype relationship. In terms of statistical modelling, two classes of models were considered: i) Mixed models, based on genomic relationship matrix to define the (co)variance between relatives (called GBLUP model); and ii) Multilocus association models, which thousands of markers are modeled simultaneously and the marker effects are summed, in order to compute the genetic merit of individuals. Both approaches were considered in separated chapters. Chapter entitled \"A mixed model to multiplicative harvest-location trial applied to genomic prediction in Coffea canephora\" addressed an expansion of the traditional GBLUP to accommodate interaction effects (Genotype × Local and Genotype × Harvest). For this end, we have tested appropriate (co)variance structures for modeling heterogeneity and correlation of genetic effects and residual effects. The proposed model, called MET.GBLUP, showed the best goodness of fit and higher predictive ability, when compared to other methods. Chapter in the sequence was entitled \"Comparison of statistical methods and reliability of genomic prediction in Coffea canephora population\" and addressed the use of different modelling assumptions considering multilocos association models. The usual assumption of marker effects drawn from a normal distribution was relaxed, in order to seek for a possible dependency between predictive performance and trait, conditional on the genetic architecture. Although the competitor models are conceptually different, a minimal difference in predictive accuracy was observed in the comparative analysis. In terms of computational demand, Bayesian models showed higher time of analysis. Results discussed in both chapters have supported the potential of genomic selection to reshape traditional breeding programs. In practice, compared to traditional phenotypic evaluation, it is expected to accelerate the breeding cycle in recurrent selection programs, maintain genetic diversity and increase the genetic gain per unit of time.