Análise genômica por janelas cromossômicas: regressão funcional e saltos reversíveis

• Autor(a) principal: Moura, Ernandes Guedes
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/49422
• topic: B-Splines; Modelos funcionais; Séries de Fourier; Markov Chain Monte Carlo; Functional models; Fourier series; Estatística

There are many methods for genomic selection that address the problems of multicollinearity and high dimensionality, among which the rr-BLUP and Bayes B stand out in the literature. Methods of continuous genome and functional regression in chromosomal windows (bins) were recently proposed to better utilize the linkage disequilibrium between SNP (Single Nucleotide Polymorphism) and potential QTLs (Quantitative Trait Loci). One of the proposed strategies is to use polynomial or trigonometric functions in bins-fitted versions. In this case, a complicating factor is the potential misspecification of the number and the sizes of the bins, with a potential increase in the prediction error. In this thesys we investigate the advantages of making inference in the joint posterior distribution for the number, the size and the effects of marks in bins in a reversible jump sampling process. This type of technique was difficult to implement for previous models that took into account the distance between marks and QTLs, but it can be greatly simplified in simple regression models typical of modern genomics (where each SNP potentially segregates as a QTL). We study the two strategies and their immediate consequences for genomic selection. A basic review of the literature methods was used to subsidize two original papers. In the first one, we evaluated the implementation of functional models of bins using Fourier series and B- Splines. In the second, we introduce the RJ-MCMC (Reverse Jump Markov Chain Monte Carlo) for functional models in which each bin is represented in the sampling by only one of its marks. The models considered were comparable to the most used for prediction (Bayes-B, rr-BLUP) and are suitable for genomic selection. As a potential by-product of the thesys, the results for association studies are also interesting, despite not being our main goal to evaluate them.