The overall goal of this thesis was to improve the accuracy of genomic prediction in dairy and beef cattle by developing, evaluating and enhancing novel or existent models and approaches for genomic selection. Four studies were conducted to fulfill this goal. In the first study, the impact of using genotypes imputed from low density panels for genomic prediction was evaluated and compared between a Bayesian mixture model and the Genomic Best Linear Unbiased Prediction (GBLUP) method. Results showed that for traits affected by a few large QTL, the Bayesian mixture model resulted in greater reduction in accuracy of genomic prediction, compared to GBLUP. However, for all SNP panels, scenarios and all traits studied, the Bayesian mixture model produced greater or similar accuracy, compared to the GBLUP method. In the second study, a new computing algorithm, called right-hand side updating strategy (RHSU), was proposed and compared to the conventional Gauss-Seidel residual update algorithm (GSRU) for genomic prediction. Results showed that RHSU would outperform GSRU once the sample size exceeded a fraction of the number of the SNPs. As the sample size continued to grow, the RHSU algorithm became more efficient than GSRU. In the third study, three different strategies of forming a training population for genomic prediction, within-breed, across-breed and pooling data from different breeds, were evaluated in Angus and Charolais steers using phenotypes on residual feed intake (RFI) and genotypes on the Illumina BovineSNP50 Beadchip (50k). Results suggested that using the 50k SNP panel, within-breed genomic prediction was a safe strategy; across-breed prediction resulted in the lowest accuracy; pooling data from different breeds had a potential to improve the accuracy but should be conducted with caution due to possible loss of accuracy. In the last study, a multi-task Bayesian learning model was proposed for multi-population genomic prediction. The performance of the multi-task model was evaluated in Holstein and Ayrshire dairy breeds. Results showed that the multi-task Bayesian learning model is effective and could be beneficial to smaller populations where only a limited number of training animals are available.

This work focuses on the evaluation of a bagging EB method in terms of its ability to select a subset of QTL-related markers for accurate EBV prediction. Experiments were performed on several simulated and real datasets consisting of SNP genotypes and phenotypes. The simulated datasets modeled different dominance levels and different levels of background noises. Our results show that the bagging EB method is able to detect most of the simulated QTL, even with large background noises. The average recall of QTL detection was $0.71$. When using the markers detected by the bagging EB method to predict EBVs, the prediction accuracy improved dramatically on the simulation datasets compared to using the entire set of markers. However, the prediction accuracy did not improve much when doing the same experiments on the two real datasets. The best accuracy of EBV prediction we achieved for the dairy dataset is 0.57 and the best accuracy for the beef dataset is 0.73.

No description available.

Doctor of Philosophy / Department of Agronomy / Allan Fritz / Wheat breeders are constantly working to develop new wheat varieties with improved performance for agronomically important traits such as yield and disease resistance. Identifying better ways of phenotyping germplasm, developing methods for predicting performance based on genetic information, and identifying novel sources of genetic disease resistance can all improve the efficiency of breeding efforts. Three studies relating to these research interests were conducted. Synthetic hexaploid wheat lines were screened for resistance to root-lesion nematodes, an economically important pest of wheat. This resulted in the identification of three lines resistant to the root-lesion nematode species Pratylenchus thornei. Grain yield data from multi-location yield trials and average yields for counties in Kansas were used to identify wheat production areas in Kansas. Knowledge obtained from this study is useful for both interpreting data from yield trials and deciding where to place them in order to identify new higher yielding varieties. These data also aided the final research study, developing a genomic selection (GS) model for yield in the Kansas State University wheat breeding program. This model was used to assess the accuracy of GS in conditions experienced in a breeding project. Available measurements of GS have been constructed using simulations or using conditions not typical of those experienced in a wheat breeding program. The estimate of accuracy determined in this study was less than many of the reported measurements. This measure of accuracy will aid in determining if GS is a cost efficient tool for use in wheat breeding.

Genomic selection (GS) has been adopted by the dairy cattle breeding industry and the opportunity exists to implement this technology in UK beef and sheep breeding. However, these sectors do not appear so readily predisposed to GS implementation. Following an introduction to GS in Chapter 1, Chapter 2 investigated the structure of the little-studied UK beef breeding sector. This provided estimates of key commercial and pedigree population parameters, for use in modelling genetic gain from GS. Terminal traits were found to be the dominant selection goals, with 85% of beef-sired commercial matings resulting in progeny being slaughtered at a prime age. Pedigree bulls disseminated the majority of genes in the sector via natural service. The correlation between the terminal selection index (TI) and the sale price of breeding bulls was moderate, suggesting a modest uptake of genetic technology in the sector. Chapter 3 estimated selection intensity for TI, generation interval and the dissemination rate of improved genes in the pedigree Limousin population. In order to predict the genetic gain achievable in using GS in beef and sheep breeding, Chapters 4 to 6 undertook deterministic selection index simulations, incorporating genomic information as correlated traits. In Chapter 4, GS was modelled for terminal beef traits, when incorporating carcass information and accounting for likely genotype by environment interaction. Using a training population of 2000 sires, this concept was predicted to offer 40% greater genetic gain than existing BLUP selection using pedigree phenotypes. Gene flow methodology projected the commercial value of this gain to offer a substantial return net of genotyping costs. Chapter 5 explored GS for maternal beef traits within the concept of a nucleus breeding scheme. Whilst greater genetic gain was predicted with GS than with conventional BLUP, the economic value of this gain was projected to be too low to justify such a scheme in the UK. Chapter 6 proposed a synergy between computer tomography (CT) phenotypes and GS in sheep breeding. Developing a genomic predictor from male selection candidates with CT phenotypes and conventional performance records was predicted to increase genetic gain by 55% over BLUP selection without CT traits. Introducing GBV contributed most of the accuracy in this scenario, suggesting that the existing performance recording structure in UK sheep breeding could in the future be replaced by GS using CT. In the general discussion, the potential for GS in other beef and sheep traits was considered in the light of the outcomes of these simulations. Given the lack of vertical integration in UK beef and sheep sectors, the drivers for implementation of GS are examined. Finally, the options for international cooperation and the possibilities offered by future genotyping technology are considered. It was concluded that implementation of GS incorporating beef carcass phenotypes was merited and could provide a platform for future GS implementation in other novel traits. Sheep GS with CT traits was considered a higher risk strategy, due to the lack of evidence for uptake of existing EBV technology.

Doctor of Philosophy / Genetics Interdepartmental Program - Plant Pathology / Jesse A. Poland / A major challenge for 21st century plant geneticists is to predict plant performance based on genetic information. This is a daunting challenge, especially when there are thousands of genes that control complex traits as well as the extreme variation that results from the environment where plants are grown. Rapid advances in technology are assisting in overcoming the obstacle of connecting the genotype to phenotype. Next generation sequencing has provided a wealth of genomic information resulting in numerous completely sequenced genomes and the ability to quickly genotype thousands of individuals. The ability to pair the dense genotypic data with phenotypic data, the observed plant performance, will culminate in successfully predicting cultivar performance. While genomics has advanced rapidly, phenomics, the science and ability to measure plant phenotypes, has slowly progressed, resulting in an imbalance of genotypic to phenotypic data. The disproportion of high-throughput phenotyping (HTP) data is a bottleneck to many genetic and association mapping studies as well as genomic selection (GS). To alleviate the phenomics bottleneck, an affordable and portable phenotyping platform, Phenocart, was developed and evaluated. The Phenocart was capable of taking multiple types of georeferenced measurements including normalized difference vegetation index and canopy temperature, throughout the growing season. The Phenocart performed as well as existing manual measurements while increasing the amount of data exponentially. The deluge of phenotypic data offered opportunities to evaluate lines at specific time points, as well as combining data throughout the season to assess for genotypic differences. Finally in an effort to predict crop performance, the phenotypic data was used in GS models. The models combined molecular marker data from genotyping-by-sequencing with high-throughput phenotyping for plant phenotypic characterization. Utilizing HTP data, rather than just the often measured yield, increased the accuracy of GS models. Achieving the goal of connecting genotype to phenotype has direct impact on plant breeding by allowing selection of higher yielding crops as well as selecting crops that are adapted to local environments. This will allow for a faster rate of improvement in crops, which is imperative to meet the growing global population demand for plant products.

The management of genomic diversity is important within breeding programs and is primarily achieved through controlling the rate of inbreeding. A failure to adequately manage the rate of inbreeding will result in an increased risk of the expression of lethal recessive mutations, inbreeding depression and losses in genetic variance, thereby restricting long-term genetic progress. Each research chapter within this thesis used real data collected from a commercial pig breeding operation to examine a key area of research regarding the management of genomic diversity. The first research chapter examined the selection outcomes from the practical application of Optimal Contributions (OC). These outcomes were examined to determine their alignment with the current theories regarding selection, particularly as to the extent by which selection decisions were influenced by estimated Mendelian sampling terms. This assessment was conducted for the initial selection of individuals as parents, which parents went on to provide a long-term contribution and the magnitude of these contributions. OC was shown to have shifted breeding decisions more closely in alignment with the estimated Mendelian sampling terms. The second research chapter used genomic data to assess the adequacy of the pedigree-based approach for managing diversity during selection. This approach assumes the infinitesimal model with all loci neutral and no impact from selection per se on heterozygosity. Using genomic information, the observed loss of heterozygosity at each marker was compared to the loss of heterozygosity expected from the pedigree-based relationships. Regional disparities between the observed and expected losses in heterozygosity were detected, which were potentially attributable to selection. Runs of homozygosity and the pairwise linkage disequilibrium between markers were also examined within these regions. Regions showing disparity were found to contain well validated quantitative trait loci for important traits. The third research chapter sought to provide a genomic solution to the shortcomings of the pedigree-based approach for quantifying relatedness, identified above. A methodology was devised for tracing identity by descent (IBD) at each allelic position over five ancestral generations, following phasing and imputation of the genomic data. A comparison was made between the inbreeding expected from the pedigree relationships and that observed from the identity by descent of genomic information. In the population studied it was not currently feasible to derive a relationship matrix based exclusively on observed IBD. The fourth research chapter used imputed genomic information to identify haplotypes which had a putative lethal recessive effect. Haplotypes which were never observed in the homozygous form, either in the population or in the offspring produced between carriers, were classified as candidate haplotypes. The top candidates on each chromosome were then examined for a reduction in the total number born when two carriers were mated together. A total of six putative lethal recessive haplotypes were detected relating to at least four putative lethal recessive mutations, where one homozygote was absent and the size of the reduction in litter size matched that expected for a lethal recessive effect. The research chapters contained within this thesis demonstrate the important role that genomics can have in managing inbreeding in addition to generating genetic gain. Genomics is able to provide a more accurate prediction of the Mendelian sampling term, better quantify the relatedness between individuals and detect lethal recessive effects.

Genomic evaluation techniques have been a huge success in the dairy cattle industry, as they allow accurate enough estimation of breeding values at a young age to allow selection decisions to be made at an earlier stage, thereby increasing the rate of genetic progress per annum. The success of genomic selection techniques relies on the existence of linkage disequilibrium (LD) between markers and quantitative trait loci (QTL) across the population of interest; LD persists across larger distances within breeds than across breeds. Therefore, most success so far has been for selection within breeds, but the industry is keen for “across breed” evaluations to be developed, both in a multi-breed scenario which would allow evaluations for breeds that are numerically too small to carry out evaluations within breeds, and also for the evaluation of crossbred animals. This thesis investigates the potential for applying genomic selection techniques in both the multi-breed and crossbred scenarios. Chapter 2 examines the potential for a multi-breed reference population to improve the accuracy of genomic evaluation for a numerically small breed, for a range of production and non-production traits. The results provide evidence that forming a multi-breed reference population for two closely related breeds (Holstein and Friesian) results in a higher accuracy of GEBVs for the smaller breed, particularly when more phenotypic records are added via the single-step GBLUP method, and when a higher density SNP chip is used. Chapter 3 examines the crossbred scenario, whereby GEBVs are calculated for crossbred individuals based on a crossbred reference population. The population used for analysis was a highly crossbred African population, and GEBVs were calculated for three groups of animals chosen according to whether they had a high or low proportion of imported dairy genetics. Accuracy of prediction was higher than expected, and provided proof of concept for applying genomic selection techniques in crossbred African cattle populations. Chapter 4 investigates the potential for using novel SNPs derived from sequence data in order to estimate genomic relationships across cattle breeds, deploying data from two closely related breeds, Fleckvieh and Simmental, and a further distant European breed, the Brown Swiss. Novel SNPs were selected from sequence based on their putative impact on the genome, with impacts being inferred by SNP annotation software snpEff. Results showed that genomic relationships calculated using novel SNPs have a high correlation with genomic relationships calculated using SNPs common to the Illumina BovineHD SNP chip, though between-breed correlations were lower than those within breeds. The results presented in this thesis demonstrate that utilising a multi-breed reference population can improve the accuracy of prediction for a numerically small breed, and that genomic prediction of highly crossbred individuals is also feasible. However, differences between breeds and also types of crossbred animal suggest that no one solution can be used for all across-breed evaluations, and further research will be needed to allow commercial implementation in further populations.

Quantitative genetics is the study of continuously distributed traits and their ge- netic components. Recent developments in DNA sequencing technologies and computational systems allow researchers to conduct large scale in silico studies. However, going from raw DNA reads to genomic prediction of quantitative traits with the help of neural networks is a long and error-prone process. In the course of this thesis, many steps involved in this process will be assessed in depth. Chap- ter 2 will feature a study that compares the landscape of chloroplast genome as- sembly tools. Chapter 3 will present a software to perform genome-wide associa- tion studies using modern tools, which allow GWAS-Flow to outperform current state of the art software packages. Chapter 4 will give an in depth introduc- tion to machine learning and the nature of quantitative traits and will combine those to genomic prediction with artificial neural networks and compares the re- sults to those of algorithms based on linear mixed models. Finally, in Chapter 5 the results from the previous chapters are summarized and used to elucidate the complex nature of studies concerning quantitative genetics. / Quantitative Genetik beschäftigt sich mit kontinuierlich verteilten Merkmalen und deren genetischer Komponenten. In den letzten Jahren gab es vielfältige Entwicklungen in der Computertechnik und der Genomik, insbesondere der DNA Sequenzierung, was Forschern erlaubt großflächig angelegte in silico Studien durchzuführen. Jedoch ist es ein komplexer Prozess von rohen Sequenzdaten bis zur genomischen Vorhersage mit Hilfe von neuronalen Netzwerken zu kommen. Im Rahmen der vorliegenden Studien werden viele Schritte, die an diesem Prozess beteiligt sind beleuchtet. Kapitel 2 wird einen Vergleich zwischen einer Vielzahl an Werkzeugen zur Assemblierung von Chloroplasten Genomen ziehen. Kapitel 3 stellt eine neu entwickelte Software zur genom-weiten Assoziationskartierung vor, die bisherigen Programmen überlegen ist. Kapitel 4 stellt maschinelles Lernen und die genetischen Komponenten von quantitativen Merkmalen vor und bringt diese im Kontext der genomischen Vorhersagen zusammen. Zum Schluss in Kapitel 5 werden die vorherigen Ergebnisse im Gesamtkontext der quantitativen Genetik erläutert.

Doctor of Philosophy / Genetics Interdepartmental Program / Jesse Poland / The Green Revolution is credited with saving billions of lives by effectively harnessing new genetic resources and breeding strategies to create high-yielding varieties for countries lacking adequate food security. To keep the next billion people in a state of food security, plant breeders will need to rapidly incorporate novel approaches and technologies into their breeding programs. The work presented here describes new genomic and phenomic strategies and tools aimed at accelerating genetic gain in plant breeding. Plant breeders have long relied on regional testing networks to evaluate new breeding lines across many locations. These are an attractive resource for both retrospective and contemporary analysis due to the vast amount of data available. To characterize genetic progress of plant breeding programs in the Central Plains, entries from the Southern Regional Performance Nursery dating back to 1992 were evaluated in field trials. The trend for annual improvement was 1.1% yr⁻¹, matching similar reports for genetic gain. During the same time period, growth of on-farm yields stagnated. Genomic selection, a promising method to increase genetic gain, was tested using historical data from the SRPN. A temporal-based model showed that, on average, yield predictions outperformed a year-to-year phenotypic correlation. A program-based model found that the predictability of a breeding program was similar when using either data from a single program or from the entire regional collection. Modern DNA marker platforms either characterize a small number of loci or profile an entire genome. Spiked genotyping-by-sequencing (sGBS) was developed to address the need in breeding programs for both targeted loci and whole-genome selection. sGBS uses a low-cost, integrated approach that combines targeted amplicons with reduced representation genotyping-by-sequencing. This approach was validated using converted and newly-designed markers targeting known polymorphisms in the leaf rust resistance gene Lr34. Plant breeding programs generate vast quantities of data during evaluation and selection of superior genotypes. Many programs still rely on manual, error-prone methods to collect data. To make this process more robust, we have developed several open-source phenotyping apps with simple, intuitive interfaces. A contemporary Green Revolution will rely on integrating many of these innovative technologies into modern breeding programs.