Bruce Golden

Producing and using genetic evaluations in the United States beef industry of today.

The overall motivation for the development of an information system for beef cattle improvement is the belief that knowledge of breeding values and heterosis effects allows one to determine the consequences of alternative selection and mating options. With this information, livestock managers can easily shift populations in a desirable direction. The foundation principles for establishing a sound breeding program, including the prediction of animal performance for economically relevant traits and their incorporation into a single index of aggregate economic merit, have been well established over the last half century. Rather than this goal-based approach, the industry adopted a data-driven approach to the production of genetic evaluations that has been characterized by an overemphasis on the evaluation of productive traits, notably BW at various ages, with inadequate regard for other economically important traits, such as reproduction, animal health, and feed requirements. Production of evaluations is breed association centered, and this has delayed the introduction of national across-breed evaluations for all breeds and crosses of cattle. The computational aspects of producing evaluations are now migrating from land-grant universities to breed associations, but not yet to a single entity. The introduction of genomic information in the form of high-density SNP panels will introduce threats, challenges, and new opportunities for the production of evaluations, and represents the largest force to alter the structure of the beef improvement industry since the advent of AI. The use of evaluations has, until recently, stopped short of the provision of index merit as a basis for selection. Accordingly, the value propositions associated with annual improvement or the selection of alternative sires has not been well communicated. Technology, along with economic and other issues related to stakeholder acceptance, will collectively determine the future nature of the industry in terms of the production and use of evaluations.

Milestones in beef cattle genetic evaluation

National beef cattle genetic evaluation programs have evolved in the United States over the last 35 yr to create important tools that are part of sustainable breeding programs. The history of national beef cattle genetic evaluation programs has lessons to offer the next generation of researchers as new approaches in molecular genetics and decision support are developed. Through a series of complex and intricate pressures from technology and organizational challenges, national cattle evaluation programs continue to grow in importance and impact. Development of enabling technologies and the interface of the disciplines of computer science, numerical methods, statistics, and quantitative genetics have created an example of how academics, government, and industry can work together to create more effective solutions to technical problems. The advent of mixed model procedures was complemented by a series of breakthrough discoveries that made what was previously considered intractable a reality. The creation of modern genetic evaluation procedures has followed a path characterized by a steady and constant approach to identification and solution for each technical problem encountered. At its core, the driving force for the evolution has been the need to constantly improve the accuracy of the predictions of genetic merit for breeding stock, especially young animals. Sensible approaches, such as the principle of economically relevant traits, were developed that created the rules to be followed as the programs grew. However, the current systems are far from complete or perfect. Modern genetic evaluation programs have a long way to go, and a great deal of improvement in the accuracy of prediction is still possible. But the greatest challenge remains: the need to understand that genetic predictions are only parameters for decision support procedures and not an end in themselves.

Computational strategies for alternative single-step Bayesian regression models with large numbers of genotyped and non-genotyped animals.

BackgroundTwo types of models have been used for single-step genomic prediction and genome-wide association studies that include phenotypes from both genotyped animals and their non-genotyped relatives. The two types are breeding value models (BVM) that fit breeding values explicitly and marker effects models (MEM) that express the breeding values in terms of the effects of observed or imputed genotypes. MEM can accommodate a wider class of analyses, including variable selection or mixture model analyses. The order of the equations that need to be solved and the inverses required in their construction vary widely, and thus the computational effort required depends upon the size of the pedigree, the number of genotyped animals and the number of loci.TheoryWe present computational strategies to avoid storing large, dense blocks of the MME that involve imputed genotypes. Furthermore, we present a hybrid model that fits a MEM for animals with observed genotypes and a BVM for those without genotypes. The hybrid model is computationally attractive for pedigree files containing millions of animals with a large proportion of those being genotyped.ApplicationWe demonstrate the practicality on both the original MEM and the hybrid model using real data with 6,179,960 animals in the pedigree with 4,934,101 phenotypes and 31,453 animals genotyped at 40,214 informative loci. To complete a single-trait analysis on a desk-top computer with four graphics cards required about 3xa0h using the hybrid model to obtain both preconditioned conjugate gradient solutions and 42,000 Markov chain Monte-Carlo (MCMC) samples of breeding values, which allowed making inferences from posterior means, variances and covariances. The MCMC sampling required one quarter of the effort when the hybrid model was used compared to the published MEM.ConclusionsWe present a hybrid model that fits a MEM for animals with genotypes and a BVM for those without genotypes. Its practicality and considerable reduction in computing effort was demonstrated. This model can readily be extended to accommodate multiple traits, multiple breeds, maternal effects, and additional random effects such as polygenic residual effects.

Comparison of alternative approaches to single-trait genomic prediction using genotyped and non-genotyped Hanwoo beef cattle

BackgroundGenomic predictions from BayesA and BayesB use training data that include animals with both phenotypes and genotypes. Single-step methodologies allow additional information from non-genotyped relatives to be included in the analysis. The single-step genomic best linear unbiased prediction (SSGBLUP) method uses a relationship matrix computed from marker and pedigree information, in which missing genotypes are imputed implicitly. Single-step Bayesian regression (SSBR) extends SSGBLUP to BayesB-like models using explicitly imputed genotypes for non-genotyped individuals.MethodsCarcass records included 988 genotyped Hanwoo steers with 35,882 SNPs and 1438 non-genotyped steers that were measured for back-fat thickness (BFT), carcass weight (CWT), eye-muscle area, and marbling score (MAR). Single-trait pedigree-based BLUP, Bayesian methods using only genotyped individuals, SSGBLUP and SSBR methods were compared using cross-validation.ResultsMethods using genomic information always outperformed pedigree-based BLUP when the same phenotypic data were modeled from either genotyped individuals only or both genotyped and non-genotyped individuals. For BFT and MAR, accuracies were higher with single-step methods than with BayesB, BayesC and BayesCπ. Gains in accuracy with the single-step methods ranged from +0.06 to +0.09 for BFT and from +0.05 to +0.07 for MAR. For CWT, SSBR always outperformed the corresponding Bayesian methods that used only genotyped individuals. However, although SSGBLUP incorporated information from non-genotyped individuals, prediction accuracies were lower with SSGBLUP than with BayesC (πxa0=xa00.9999) and BayesB (πxa0=xa00.98) for CWT because, for this particular trait, there was a benefit from the mixture priors of the effects of the single nucleotide polymorphisms.ConclusionsSingle-step methods are the preferred approaches for prediction combining genotyped and non-genotyped animals. Alternative priors allow SSBR to outperform SSGBLUP in some cases.

Genetic parameters for carcass and ultrasound traits in Hereford and admixed Simmental beef cattle: Accuracy of evaluating carcass traits1

Genetic parameters are required to evaluate carcass merit using correlated real-time ultrasound (RTU) measurements. Many registered bulls and heifers are measured using RTU before consideration for selection as parents, whereas few animals are recorded for carcass traits and those are often crossbred steers. The objective of this study was to estimate genetic parameters required for evaluating carcass merit in the American Hereford Association (AHA) and the American Simmental Association (ASA) using multivariate models and to assess accuracy of carcass trait estimated breeding values (EBV) for selection candidates. All available carcass data including carcass weight (CWT), fat thickness (FAT), longissimus muscle area (LMA), and marbling score (MRB) were provided by the AHA and the ASA along with RTU data including fat thickness (UFAT), longissimus muscle area (ULMA), and percentage of intramuscular fat (UIMF). Carcass data comprised 6,054 AHA and 9,056 ASA cattle, while RTU data in comparable numbers from close relatives comprised 6,074 AHA and 7,753 ASA cattle. Pedigrees included 33,226 AHA and 37,665 ASA animals. Fixed effects for carcass and RTU data included contemporary group, age at scan/slaughter, and major breed percentages. Restricted maximum likelihood procedures were applied to all the carcass and RTU measurements, along with birth weight to account for selection, fitting 8-trait multivariate models separately for each breed association. Heritability estimates for AHA and ASA carcass traits were 0.41 ± 0.04 and 0.25 ± 0.03 for FAT, 0.47 ± 0.04 and 0.32 ± 0.03 for LMA, 0.48 ± 0.04 and 0.43 ± 0.04 for MRB, 0.51 ± 0.04 and 0.34 ± 0.03 for CWT, and for RTU traits were 0.29 ± 0.04 and 0.37 ± 0.03 for UFAT, 0.31 ± 0.04 and 0.44 ± 0.03 for ULMA, and 0.45 ± 0.04 and 0.42 ± 0.03 for UIMF. Genetic correlations for AHA and ASA analyses between FAT and UFAT were 0.74 ± 0.08 and 0.28 ± 0.13, between LMA and ULMA were 0.81 ± 0.07 and 0.57 ± 0.10, and between MRB and UIMF were 0.54 ± 0.08 and 0.73 ± 0.07. Predictions of carcass merit using RTU measurements in Hereford cattle would be more reliable for FAT and LMA than MRB, but the reverse would be true for admixed Simmental cattle. Genetic correlations for MRB in AHA and for FAT and LMA in ASA are less than currently assumed in their national evaluations. Collection of greater numbers of carcass measurements would improve the accuracy of genetic evaluations for carcass traits in both breeds.

Estimation of Genetic Parameters for Carcass Traits and Their Corresponding Ultrasound Measurements in Crossbred Beef Cattle

and Implications Variance parameters including heritabilities, genetic and residual correlations are required for national cattle evaluation. There are huge amounts of data available for estimating such variance parameters for growth traits, but much less data is available for carcass traits. In this study, heritabilities and genetic correlations were estimated using restricted maximum likelihood on carcass weight (CWT), fat thickness (FAT), longissimus muscle area (LMA), marbling score (MRB), birth weight (BW), and ultrasound measurements of fat thickness (UFAT), longissimus muscle area (ULMA) and estimated percentage of intramuscular fat (UIMF) for crossbred cattle with carcass data recorded by the American Simmental Association. A multivariate animal model was fitted using ASREML4 software. The results demonstrate that UIMF measurements provide some useful information for carcass MRB (rg=0.73), but genetic correlations were only moderate between ULMA and LMA (0.56) and were weak between UFAT and FAT (0.38). The implications are that carcass measurements on progeny are the most reliable approach to evaluate carcass traits. Introduction American Simmental Association (ASA) like other breed associations has long been using real time ultrasound data in addition to carcass measurements to enhance national cattle evaluations on carcass traits. Studies revealed that evaluations combining ultrasound and carcass data outperform the ones that are based on carcass data alone. To date, most reports of parameters for use in national cattle evaluation were based on bivariate animal model analyses and some have fitted contemporary groups of fixed breed fractions rather than accounting for breed percentages. The objective of this study was to estimate genetic parameters required for evaluating carcass merit in the multi-breed analyses undertaken by International Genetic Solutions, using a single multivariate model fitting major breed percentages as fixed effects.

A heterogeneous compute solution for optimized genomic selection analysis

This paper presents a heterogeneous computing solution for an optimized genetic selection analysis tool, GenSel. GenSel can be used to efficiently infer the effects of genetic markers on a desired trait or to determine the genomic estimated breeding values (GEBV) of genotyped individuals. To predict which genetic markers are informational, GenSel performs Bayesian inference using Gibbs sampling, a Markov Chain Monte Carlo (MCMC) algorithm. Parallelizing this algorithm proves to be a technically challenging problem because there exists a loop carried dependence between each iteration of the Markov chain. The approach presented in this paper exploits both task-level parallelism (TLP) and data-level parallelism (DLP) that exists within each iteration of the Markov chain. More specifically, a combination of CPU threads using OpenMP and GPU threads using NVIDIAs CUDA paradigm is implemented to speed up the sampling of each genetic marker used in creating the model. Performance speedup will allow this algorithm to accommodate the expected increase in observations on animals and genetic markers per observation. The current implementation executes 1.84 times faster than the optimized CPU implementation.