Year: 2013
Type: dissertation
Abstract: Genome-wide marker data is used in animal and plant breeding in computing genomic breeding values, and in human genetics in identifying disease susceptibility genes, predicting unobserved phenotypes and assessing disease risks. While the tremendous number of markers available for easy and cost-effective genotyping is an invaluable asset in genetic research and in animal and plant breeding, the ever increasing data sets are placing heavy demands on the statistical analysis methodology. The statistical methods proposed for genomic selection are based on either traditional best linear unbiased prediction (BLUP) or on different Bayesian multilocus association models. The marker based genomic best linear unbiased prediction (G-BLUP) employs the marker information in estimating genomic relationships between the individuals, and then utilizes the marker-estimated genomic relationship matrix in a mixed model context. A multilocus association model, on the other hand, uses the marker information directly by assigning different effects to the marker alleles, and quantifies the genomic breeding value of an individual as the sum of the marker effects. The advantage of a multilocus association over the G-BLUP is that the former allows the estimated effect size to vary over the set of markers, while the latter assumes a constant impact throughout the genome. In human genetics the most prevalent approach is a single SNP association model. These models consider only one marker at the time, ignoring the possible effects of the other major loci. This is less than ideal in genome-wide study for a complex trait, as such traits are assumed to be affected by a multitude of genes. The objective of this work is to obtain further understanding of the behavior of the different Bayesian multilocus association models and of the instances in which different methods work best, to seek connections between the different Bayesian models, and to develop a Bayesian multilocus association model framework, along with an efficient parameter estimation machinery, that can be utilized in phenotype prediction, genomic breeding value estimation and quantitative trait locus (QTL) location and effect estimation from a variety of genome-wide data.
Source: Työväentutkimus Vuosikirja
Author: Hanni Kärkkäinen
Language: English
Subfield: Genetics
Domain: Life Sciences
Open Access status: green
Funders: Helsingin Yliopisto, Tekes
OpenAlex ID: W2105210732