Análise de associação genômica ampla com múltiplas marcas: seleção e associação genômica ampla unificadas

Abstract

The Genome-Wide Selection (GWS) and Genome Wide Association Studies (GWAS) techniques have been routinely applied in interaction in breeding programs, especially for agronomic crops, which generates technological/scientific, economic, and human investments. Besides the efficiency of both techniques in selecting and detecting genes responsible for specific phenotypic characteristics, the GWS is an experimental approach that integrates quantitative genetics tools and emphasizes the prediction of the genetic effects of thousands of markers (SNPs). The purpose of the GWAS is to identify variations in the genome-wide DNA sequence associated with specific interesting phenotypic characteristics. The present work intends to unify both breeding programs strategic fronts (GWS and GWAS) in the same conceptual treatment. To do so, we implemented the Genome-Wide Selection Association Studies (GWSAS) using the mixed linear model to identify candidate regions for genes expressing a specific characteristic, considering the structure populational. One of the main differences between the methods used to estimate parameters and the method required for GWSAS is the characterization of the behavior of the conditional a posteriori distribution functions obtained by eliminating disturbing parameters. Therefore, we integrated such a function with the Bayesian inference method, ensuring the capacity to model population and parentage structures based on the estimation that captures the dynamics of molecular markers (in the context of genetic micro-information). In other words, the genomic covariance structure of the nucleus of the marginally conditional a posteriori distribution allowed us to raise the dependence and correlation information between the molecular markers, and intrinsically apprehend the identity by state (IBS) and identity by descent (IBD). We also proposed a solution for inverting the V k−1 matrix, changing variables and vector products to increase the computational facilities when using the MCMC techniques. Furthermore, we applied an analysis to formalize the relationship between the GWS (classical methods used: RR-BLUP, Bayes A, Bayes B, and BSSV) and the GWAS (based on the model proposed by (YU et al., 2006) with the GWSAS (methods used: modified RR-BLUP, modified Bayes A, modified Bayes B, and modified BSSV). Thus, we explored the primary potential aspects of the GWSAS as an attractive tool for genome selection and association by applying it to 230 maize genotyped lines with 23,153 markers to identify the results of the modified methods in genes associated to interesting characteristics, elucidating the statistical properties and adaptations of each one, through graphical comparisons and the Wald test. This methodology allowed us to evaluate the performances by using simulated and real data, while considering the Sum of Squares of the Predicted Residual Values (PRESS) for heritability and the correlations between the real and predicted genetic values. The results were satisfactory concerning the ability to detect genes and low rate of false positives, also revealing better indexes during the selection even with different patterns of contraction. The modified methods and, consequently, the simulated effects of the SNPs, clearly reflected their respective selection method profiles, thus being considered reliable. We propose the analysis of recent multi-loci genomic association studies, the denominated FASTmrMLM and FASTmrEMMA algorithms, to verify the ability to detect previously simulated and real genes, compared to all other GWSAS methods.