Genomic information for breed determination, multibreed evaluation, and estimation of variance components in large populations

• Ano de defesa: 2018
• Autor(a) principal: Junqueira, Vinícius Silva
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Bovinos de corte - Genética; Algoritmos genéticos; Variação genética; Polimorfismo de nucleotídeo único; Genética e Melhoramento dos Animais Domésticos
• Link de acesso: http://www.locus.ufv.br/handle/123456789/21565

The knowledge on breed composition is of major importance under design of breeding schemes. With this respect, the estimation of such parameters must be as accurately as possible. Currently, most of genetic evaluation programs has been predicting breed composition based on pedigree datasets; but, such estimations only accounts for the expected (allele frequency) contributions across ancestors After the development and establishment of single nucleotide polymorphism (SNP) genotyping platforms on the last decade, an interest in genetic diversity studies has arisen and especially the study of individuals’ origin. The objective of the present study was to evaluate the minimum required number of ancestry informative markers necessary to differentiate Hereford, Nelore, Brahman and Braford breeds genotyped with 777 K Illumina Bovine HD Bead Chip. In addition, we also compared the effects of different panels size on breed composition inference under different AIMs methods. To that, it was used the high-density Illumina Bovine HD BeadChip with more than 777 K SNPs to elucidate the structure of Hereford, Nelore, Brahman and Braford populations. Three different ancestry informative marker methods were used to distinguish such populations. Additionally, random marker selection was considered. Admixture software was used to infer breed composition using very low-density SNP panels assembled with AIMs. Our results suggest that is possible to assign individuals to populations with high confidence using less than 8 SNP markers selected per breed. Although millions of SNP markers have been identified, only few of them are needed to accurately infer ancestry in a cost-effective manner. Pedigree information is by nature incomplete and commonly not well established simply because many of the true genetic ties existent between individuals are not a priori known or they can be even wrong. Genomic era brought new opportunities when calculating relationships between individuals. The challenge under genomic approaches is the correct definition of genetic base by the use of pedigree and genomic data. Genetic base may change as more individuals are included and are inadequately defined if populations are genetically structured. Metafounder concept relies on the definition of pseudo-individuals that describes some level of within and/or across genetic relationship between base population. The purpose of this study was to evaluate metafounder theory to estimate breeding values and the predictive ability under a single-step approach for a multibreed population. Three different scenarios were adopted to estimate variance components and to compute breeding values: pedigree-based model, single- step GBLUP and single-step GBLUP with addition of metafounders. A total of 28 different metafounders were included in the ssGBLUP+metafounder model. In general, it was possible to note that genomic models were able to greater ability to predict the future performance. Among genomic models, the inclusion of metafounder information could increment even more the predictive ability under cross-validation approach. Restricted maximum likelihood (REML) is a popular method for parameter estimation. Because it uses the mixed model equations, it is resistant to selection bias and efficient implementations are currently available. When genomic information is available, two versions of REML may be applicable. When only genotyped animals have phenotypes, genomic REML can be applied with a genomic relationship matrix. When only a fraction of animals is genotyped, a single-step REML is applicable. In general, it is of interest to include many genotyped animals in parameter estimation and into evaluations, to account for genomic selection or pre- selection. The aim of this study was to investigate to what extent generations truncation affects estimates for a simulated population under selection.The use of less generations reduced the ability of pedigree-based model in estimating the benchmark heritability (0.30). The decrease in heritabilities based on genomic information was less than using only pedigree relationships. Genomic models provided greater correlations than pedigree-based model; on average 25 points. Single-step genomic models do not require a deeper pedigree relationship to estimate reliable variance components and breeding values. The use of APY algorithm does not affect the estimation of variance components. An extra of 2 ungenotyped generations are sufficient to compute reliable variance components; as well as breeding values and accuracies.