Liuhong Chen

Accuracy of predicting genomic breeding values for residual feed intake in Angus and Charolais beef cattle.

In beef cattle, phenotypic data that are difficult and/or costly to measure, such as feed efficiency, and DNA marker genotypes are usually available on a small number of animals of different breeds or populations. To achieve a maximal accuracy of genomic prediction using the phenotype and genotype data, strategies for forming a training population to predict genomic breeding values (GEBV) of the selection candidates need to be evaluated. In this study, we examined the accuracy of predicting GEBV for residual feed intake (RFI) based on 522 Angus and 395 Charolais steers genotyped on SNP with the Illumina Bovine SNP50 Beadchip for 3 training population forming strategies: within breed, across breed, and by pooling data from the 2 breeds (i.e., combined). Two other scenarios with the training and validation data split by birth year and by sire family within a breed were also investigated to assess the impact of genetic relationships on the accuracy of genomic prediction. Three statistical methods including the best linear unbiased prediction with the relationship matrix defined based on the pedigree (PBLUP), based on the SNP genotypes (GBLUP), and a Bayesian method (BayesB) were used to predict the GEBV. The results showed that the accuracy of the GEBV prediction was the highest when the prediction was within breed and when the validation population had greater genetic relationships with the training population, with a maximum of 0.58 for Angus and 0.64 for Charolais. The within-breed prediction accuracies dropped to 0.29 and 0.38, respectively, when the validation populations had a minimal pedigree link with the training population. When the training population of a different breed was used to predict the GEBV of the validation population, that is, across-breed genomic prediction, the accuracies were further reduced to 0.10 to 0.22, depending on the prediction method used. Pooling data from the 2 breeds to form the training population resulted in accuracies increased to 0.31 and 0.43, respectively, for the Angus and Charolais validation populations. The results suggested that the genetic relationship of selection candidates with the training population has a greater impact on the accuracy of GEBV using the Illumina Bovine SNP50 Beadchip. Pooling data from different breeds to form the training population will improve the accuracy of across breed genomic prediction for RFI in beef cattle.

Phenotypic and genetic relationships of feed efficiency with growth performance, ultrasound, and carcass merit traits in Angus and Charolais steers

Feed efficiency is of particular importance to the beef industry, as feed costs represent the single largest variable cost in beef production systems. Selection for more efficient cattle will lead to reduction of feed related costs, but should not have adverse impacts on quality of the carcass. In this study, we evaluated phenotypic and genetic correlations of residual feed intake (RFI), RFI adjusted for end-of-test ultrasound backfat thickness (RFIf), and RFI adjusted for ultrasound backfat thickness and LM area (RFIfr) with growth, ultrasound, and carcass merit traits in an Angus population of 551 steers and in a Charolais population of 417 steers. In the Angus steer population, the phenotypic and genetic correlation of RFI with carcass merit traits including HCW, carcass backfat, carcass LM area, lean meat yield, and carcass marbling were not significant or weak with correlations coefficients ranging from -0.0007 ± 0.05 to 0.18 ± 0.21. In the Charolais steer population, the phenotypic and genetic correlations of RFI with the carcass merit traits were also weak, with correlation coefficients ranging from -0.07 ± 0.06 to 0.19 ± 0.18, except for the genetic correlation with carcass average backfat, which was moderate with a magnitude of 0.42 ± 0.29. Inclusion of ultrasound backfat thickness in the model to predict the expected daily DMI for maintenance explained on average an additional 0.5% variation of DMI in the Angus steers and 2.3% variation of DMI in the Charolais steer population. Inclusion of both the ultrasound backfat and LM area in the model explained only 0.7% additional variance in DMI in the Angus steer population and only 0.6% in the Charolais steer population on top of the RFIf model. We concluded that RFIf adjusted for ultrasound backfat at the end of the test will lead to decreases of both the phenotypic and genetic correlations with carcass backfat and marbling score to a greater extent for late-maturing beef breeds such as Charolais than for early-maturing beef breeds such as Angus. However, further inclusion of ultrasound LM area on top of the final ultrasound backfat in the model of calculating RFI had little effect in reducing the correlations of RFI with the carcass merit traits.

Impact of genotype imputation on the performance of GBLUP and Bayesian methods for genomic prediction.

The aim of this study was to evaluate the impact of genotype imputation on the performance of the GBLUP and Bayesian methods for genomic prediction. A total of 10,309 Holstein bulls were genotyped on the BovineSNP50 BeadChip (50 k). Five low density single nucleotide polymorphism (SNP) panels, containing 6,177, 2,480, 1,536, 768 and 384 SNPs, were simulated from the 50 k panel. A fraction of 0%, 33% and 66% of the animals were randomly selected from the training sets to have low density genotypes which were then imputed into 50 k genotypes. A GBLUP and a Bayesian method were used to predict direct genomic values (DGV) for validation animals using imputed or their actual 50 k genotypes. Traits studied included milk yield, fat percentage, protein percentage and somatic cell score (SCS). Results showed that performance of both GBLUP and Bayesian methods was influenced by imputation errors. For traits affected by a few large QTL, the Bayesian method resulted in greater reductions of accuracy due to imputation errors than GBLUP. Including SNPs with largest effects in the low density panel substantially improved the accuracy of genomic prediction for the Bayesian method. Including genotypes imputed from the 6 k panel achieved almost the same accuracy of genomic prediction as that of using the 50 k panel even when 66% of the training population was genotyped on the 6 k panel. These results justified the application of the 6 k panel for genomic prediction. Imputations from lower density panels were more prone to errors and resulted in lower accuracy of genomic prediction. But for animals that have close relationship to the reference set, genotype imputation may still achieve a relatively high accuracy.

Estimates of genetic parameters for fatty acids in brisket adipose tissue of Canadian commercial crossbred beef steers

Heritability and genetic and phenotypic correlations between 15 individuals and 10 groups of fatty acids with a concentration greater than 0.5% in the brisket adipose tissue of 223 Angus and Charolais based crossbred commercial steers were estimated using univariate and bivariate animal models. Individual saturated fatty acids were low to moderately heritable, with heritability estimates ranging from 0.05 (C16:0) to 0.31 (C15:0). Individual monounsaturated fatty acids were low to moderately highly heritable ranging from 0.04 (9c C17:1 and 11c C18:1) to 0.51 (9c C14:1). Polyunsaturated fatty acid C18:2n-6 was moderately heritable (0.17). Among groups of fatty acids, heritability estimates ranged from 0.03 for branched chain fatty acid (BCFA) and n-6/n-3 to 0.16 for n-6 and Health Index. A range of low (0.00) to high (1.00) phenotypic and genetic correlations was observed among the 25 fatty acids considered in this study. In general, fatty acids such as conjugated linoleic acid (CLA) and 11t C18:1, with potential health benefits, showed significant antagonistic correlations with unhealthy fatty acids such as C14:0 and C16:0. The results from this study provide insight into the direct genetic control of host genes on fatty acid composition of beef tissues and will facilitate designs of genetic selection and/or genetic based diet management to improve fatty acid composition in beef cattle.

Multi-population genomic prediction using a multi-task Bayesian learning model

BackgroundGenomic prediction in multiple populations can be viewed as a multi-task learning problem where tasks are to derive prediction equations for each population and multi-task learning property can be improved by sharing information across populations. The goal of this study was to develop a multi-task Bayesian learning model for multi-population genomic prediction with a strategy to effectively share information across populations. Simulation studies and real data from Holstein and Ayrshire dairy breeds with phenotypes on five milk production traits were used to evaluate the proposed multi-task Bayesian learning model and compare with a single-task model and a simple data pooling method.ResultsA multi-task Bayesian learning model was proposed for multi-population genomic prediction. Information was shared across populations through a common set of latent indicator variables while SNP effects were allowed to vary in different populations. Both simulation studies and real data analysis showed the effectiveness of the multi-task model in improving genomic prediction accuracy for the smaller Ayshire breed. Simulation studies suggested that the multi-task model was most effective when the number of QTL was small (n = 20), with an increase of accuracy by up to 0.09 when QTL effects were lowly correlated between two populations (ρ = 0.2), and up to 0.16 when QTL effects were highly correlated (ρ = 0.8). When QTL genotypes were included for training and validation, the improvements were 0.16 and 0.22, respectively, for scenarios of the low and high correlation of QTL effects between two populations. When the number of QTL was large (n = 200), improvement was small with a maximum of 0.02 when QTL genotypes were not included for genomic prediction. Reduction in accuracy was observed for the simple pooling method when the number of QTL was small and correlation of QTL effects between the two populations was low. For the real data, the multi-task model achieved an increase of accuracy between 0 and 0.07 in the Ayrshire validation set when 28,206 SNPs were used, while the simple data pooling method resulted in a reduction of accuracy for all traits except for protein percentage. When 246,668 SNPs were used, the accuracy achieved from the multi-task model increased by 0 to 0.03, while using the pooling method resulted in a reduction of accuracy by 0.01 to 0.09. In the Holstein population, the three methods had similar performance.ConclusionsResults in this study suggest that the proposed multi-task Bayesian learning model for multi-population genomic prediction is effective and has the potential to improve the accuracy of genomic prediction.

Impact of hard vs. soft wheat and monensin level on rumen acidosis in feedlot heifers.

Many feedlot finishing diets include wheat when the relative wheat prices are low. This study was conducted to examine the responses in ruminal pH and fermentation as well as site and extent of digestion from substituting soft or hard wheat for barley grain and to determine whether an elevated monensin concentration might decrease indicators of ruminal acidosis in feedlot heifers. Five ruminally cannulated beef heifers were used in a 5 × 5 Latin square with 2 × 2 + 1 factorial arrangement. Treatments included barley (10% barley silage, 86% barley, 4% supplement, with 28 mg monensin/kg DM) and diets where barley was substituted by either soft or hard wheat with either 28 or 44 mg monensin/kg diet DM. Intake of DM was not affected by grain source, whereas increasing monensin with wheat diets reduced (P < 0.02) DMI. Mean ruminal pH was lower (P < 0.04) and durations of pH < 5.8 and pH < 5.5 greater (P < 0.03) for wheat than for barley diets. However, ruminal pH was not affected by wheat type or monensin level. Total VFA concentrations were greater (P < 0.03) for wheat than barley diets with no effect of wheat type. The molar proportion of propionate was greater (P < 0.04), whereas butyrate (P < 0.01) and ratio of acetate to propionate tended to be lower (P < 0.09), with the high as compared to low level of monensin. Replacing barley with wheat in finishing diets did not affect the duodenal flow or the digestibility of OM, likely as a result of greater (P < 0.01) NDF digestion from barley offsetting the increased (P < 0.03) supply of digested starch from wheat. Feeding soft vs. hard wheat delivered a greater (P < 0.03) duodenal supply of OM and nonammonia N with no differences in total tract nutrient digestion. The increased monensin concentration decreased the flow of OM (P < 0.01), total N (P < 0.05), and microbial protein (P < 0.05) to the small intestine due to decreased DMI. These results indicated that hard and soft wheat exhibited digestive characteristics similar to barley, but ruminal pH measurements indicate that compared with barley, wheat increased the risk of ruminal acidosis. Although an increased level of monensin had limited impact on ruminal indicators of acidosis, an increase in propionate would be expected to improve efficiency of feed use by heifers fed wheat-based finishing diets.

Phenotypic and genetic relationships of feeding behavior with feed intake, growth performance, feed efficiency, and carcass merit traits in Angus and Charolais steers1

Feeding behavior traits including daily feeding duration (FD), daily feeding head down time (HD), average feeding duration per feeding event (FD_AVE), average feeding head down time per feeding event (HD_AVE), feeding frequency (FF), and meal eating rate (ER) were analyzed to estimate their phenotypic and genetic correlations with feed intake, growth performance, residual feed intake (RFI), ultrasound, and carcass merit traits in Angus and Charolais finishing steers. Heritability estimates for FD, HD, FD_AVE, HD_AVE, FF, and ER were 0.27 ± 0.09 (SE), 0.25 ± 0.09, 0.19 ± 0.06, 0.11 ± 0.05, 0.24 ± 0.08, and 0.38 ± 0.10, respectively, in the Angus population and 0.49 ± 0.12, 0.38 ± 0.11, 0.31 ± 0.09, 0.29 ± 0.10, 0.43 ± 0.11, and 0.56 ± 0.13, respectively, in the Charolais population. In both the Angus and Charolais steer populations, FD and HD had relatively stronger phenotypic (0.17 ± 0.06 to 0.32 ± 0.04) and genetic (0.29 ± 0.17 to 0.54 ± 0.18) correlations with RFI in comparison to other feeding behavior traits investigated, suggesting the potential of FD and HD as indicators in assessing variation of RFI. In general, feeding behavior traits had weak phenotypic correlations with most of the ultrasound and carcass merit traits; however, estimated genetic correlations of the feeding behavior traits with some fat deposition related traits were moderate to moderately strong but differed in magnitude or sign between the Angus and Charolais steer populations, likely reflecting their different biological types. Genetic parameter estimation studies involving feeding behavior traits in beef cattle are lacking and more research is needed to better characterize the relationships between feeding behavior and feed intake, growth, feed utilization, and carcass merit traits, in particular with respect to different biological types of cattle.

Accuracy of predicting genomic breeding values for carcass merit traits in Angus and Charolais beef cattle

Accuracy of predicting genomic breeding values for carcass merit traits including hot carcass weight, longissimus muscle area (REA), carcass average backfat thickness (AFAT), lean meat yield (LMY) and carcass marbling score (CMAR) was evaluated based on 543 Angus and 400 Charolais steers genotyped on the Illumina BovineSNP50 Beadchip. For the genomic prediction within Angus, the average accuracy was 0.35 with a range from 0.32 (LMY) to 0.37 (CMAR) across different training/validation data-splitting strategies and statistical methods. The within-breed genomic prediction for Charolais yielded an average accuracy of 0.36 with a range from 0.24 (REA) to 0.46 (AFAT). The across-breed prediction had the lowest accuracy, which was on average near zero. When the data from the two breeds were combined to predict the breeding values of either breed, the prediction accuracy averaged 0.35 for Angus with a range from 0.33 (REA) to 0.39 (CMAR) and averaged 0.33 for Charolais with a range from 0.18 (REA) to 0.46 (AFAT). The prediction accuracy was slightly higher on average when the data were split by animals birth year than when the data were split by sire family. These results demonstrate that the genetic relationship or relatedness of selection candidates with the training population has a great impact on the accuracy of predicting genomic breeding values under the density of the marker panel used in this study.

Genomic prediction of breed composition and heterosis effects in Angus, Charolais and Hereford crosses using 50K genotypes

Abstract This study examined the feasibility and accuracy of using Illumina BovineSNP50 genotypes to estimate individual cattle breed composition and heterosis relative to estimate from pedigree. First, pedigree was used to compute breed fractions for 1124 crossbred cattle. Given the breed composition of sires and dams, retained heterosis and retained heterozygosity were computed for all individuals. Second, all animals’ genotypes were used to compute individual’s genomic breed fractions by applying a cross-validation method. Average genome-wide heterozygosity and retained heterozygosity based on genomic breed fraction were computed. Lastly, accuracies of breed composition, retained heterozygosity and retained heterosis were assessed as Pearson’s correlation between pedigree- and genome-based predictions. The average breed compositions observed were 0.52 Angus, 0.23 Charolais, and 0.25 Hereford for pedigree-based prediction and 0.46, 0.26, and 0.28 for genome-based prediction, respectively. Correlations of predicted breed composition ranged from 0.94 to 0.96. Genome-based retained heterozygosity and retained heterosis from pedigree were also highly correlated (0.96). A positive association of nonadditive genetic effects was observed for growth traits reflecting the importance of heterosis for these traits. Genomic prediction can aid analyses that depend on knowledge of breed composition and serve as a reliable method to predict heterosis to improve the efficiency of commercial crossbreeding schemes.

An alternative computing strategy for genomic prediction using a Bayesian mixture model

Chen, L., Li, C. and Schenkel, F. 2015. An alternative computing strategy for genomic prediction using a Bayesian mixture model. Can. J. Anim. Sci. 95: 1-11. Bayesian methods for genomic prediction are commonly implemented via Markov chain Monte Carlo (MCMC) sampling schemes, which are computationally demanding in large-scale applications. An alternative computing algorithm, called right-hand side updating strategy (RHSU), was proposed by exploiting the sparsity feature of the marker effects in a Bayesian mixture model. The new algorithm was compared with the conventional Gauss-Seidel residual update (GSRU) algorithm by the number of floating point operations (FLOP) required in one round of MCMC sampling. The two algorithms were also compared in a Holstein data example with the training data size varying from 1000 to 10 000 and a marker density of 35 790 single nucleotide polymorphisms (SNP). Results showed that the proposed RHSU algorithm would outperform the traditional GSRU algorithm when the sample size exceeded a fraction of the number of the SNPs, which typically varied from 0.05 to 0.18 when the proportion of SNPs with no effect on the trait varied from 0.90 to 0.95. Results from the Holstein data example agreed very well with theoretical expectations. With adoption of a 50 k SNP panel and an increasing training data size, RHSU would be very useful if Bayesian methods are preferable for genomic prediction.