Both mapping quantitative trait loci (QTL) and genomic selection (GS) contribute to the genetic improvement of livestock by combining molecular and phenotypic information.

This thesis includes a study on a method to combine linkage disequilibrium (LD) and cosegregation (CS) information to fine-map QTL, a multi-locus measure of LD and its relationship with long-term accuracy of genomic estimated breeding value (GEBV), and an approach to simulate validation data by sampling according to mendelian inheritance to predict long-term accuracy of GEBV.

A gene-frequency model is proposed to fine-map QTL that combines LD and CS information, where LD information is incorporated into the conditional means and variances given marker information, and CS information is incorporated into covariances of gametic deviations of the model. Algorithms are developed to draw Bayesian inferences on this gene-frequency model (BGF). The performance of the BGF method was compared to a regression method using least squares (LSR) or the identity-by-descent (IBD) method of Meuwissen and Goddard in power to detect and precision to map a QTL. Simulations were conducted under a range of marker densities, sample sizes and sizes of QTL. When there was only LD information in the data, the BGF method had power close or equal to that of LSR, and precision higher than that of LSR. The IBD method is another method that combines LD and CS information. When there was LD and CS information in the data, the BGF method had higher power and precision than the IBD method.

A multi-locus measure of LD, Rw², is proposed to quantify the long-term accuracy of genomic estimated breeding value (GEBV). Scanning through a genome with every SNP chosen to be a surrogate QTL, its genotypes are regressed on all the remaining SNPs, but are predicted using only the surrounding SNPs within a certain length of chromosomal segment. The value of Rw² is obtained by averaging the squared correlation between the true and predicted genotypes over all surrogate QTL. The values of Rw were higher than the long-term accuracies of GEBV based on the simulation, which suggests that distant loci from the QTL have a negative impact on predicting GEBV.

Since the measure Rw² is too optimistic to predict the long-term accuracy of GS, a new approach is proposed such that effects of the distant SNPs are not disregarded. Validation data was simulated through sampling mendelian inheritance based on the ordered genotypes of a target population. To validate this approach, a simulation study based on real 30k SNP genotypes of a layer chicken population was compared to the results based on the same population with real data. Combinations of a broad spectrum of genetic architectures with different heritabilities were included in the simulation. Accuracies from real data were within the range of accuracies from simulation for the most polygenic scenario among the eight scenarios studied. Results from real GS practice in dairy cattle and layer chicken industry substantiated that the polygenic scenario well approximated the genetic architectures of many complex traits.