A comparison of accuracy validation methods for genomic and pedigree-based predictions of swine litter size traits using Large White and simulated data

Factors affecting the accuracy of genomic prediction in joint pig populations

Genomic prediction (GP) has greatly advanced animal and plant breeding over the past two decades. GP in joint populations is a feasible method to improve the accuracy of genomic estimated breeding values in small populations. However, there is still a need to understand the factors that influence GP in joint populations. This study used simulated data and real data from Duroc pig populations to examine the impact of linkage disequilibrium (LD), causal variants effect sizes (CVESs), and minor allele frequencies (MAF) of SNPs on the accuracy of genomic prediction in joint populations. Three prediction methods were used: genomic best linear unbiased prediction (GBLUP), single-step GBLUP and multi-trait GBLUP. Results from the simulated datasets showed that the accuracies of GP in joint populations were always higher than those in a single population when only LD inconsistencies existed. However, single-step GBLUP accuracy in joint populations decreased as the correlation of MAF between populations decreased, while the accuracy of GBLUP is consistently higher in joint populations than in a single population. As the correlation of CVES between populations decreased, the accuracy of both GBLUP and single-step GBLUP in joint populations declined. Analysis of real Duroc populations showed low genetic correlation, similar to the simulated relationship between the most distant populations. In most cases in Duroc populations, GP have higher accuracies in joint populations than in individual population. In conclusion, the consistency of CVES plays a more important role in multi-population GP. The genetic relatedness of the Duroc populations is so weak that the prediction accuracy of GP in joint populations is reduced in some traits. Multi-trait GBLUP is a competitive method for the joint breeding evaluation.

Using social interaction models for genetic analysis of skin damage in gilts

BACKGROUND

Skin damage is a trait of economic and welfare importance that results from social interactions between animals. These interactions may produce wound signs on the gilt's skin as a result of damage behavior (i.e., fighting), biting syndromes (i.e., tail, vulva, or ear biting), and swine inflammation and necrosis syndrome. Although current selection for traits that are affected by social interactions primarily focuses on improving direct genetic effects, combined selection on direct and social genetic effects could increase genetic gain and avoid a negative response to selection in cases of competitive behavior. The objectives of this study were to (1) estimate variance components for combined skin damage (CSD), with or without accounting for social genetic effects, (2) investigate the impact of including genomic information on the prediction accuracy, bias, and dispersion of CSD estimated breeding values, and (3) perform a single-step genome-wide association study (ssGWAS) of CSD under a classical and a social interaction model.

RESULTS

Our results show that CSD is heritable and affected by social genetic effects. Modeling CSD with social interaction models increased the total heritable variance relative to the phenotypic variance by three-fold compared to the classical model. Including genomic information increased the prediction accuracy of direct, social, and total estimated breeding values for purebred sires by at least 21.2%. Bias and dispersion of estimated breeding values were reduced by including genomic information in classical and social interaction models but remained present. The ssGWAS did not identify any single nucleotide polymorphism that was significantly associated with social or direct genetic effects for CSD.

CONCLUSIONS

Combined skin damage is heritable, and genetic selection against this trait will increase the welfare of animals in the long term. Combined skin damage is affected by social genetic effects, and modeling this trait with a social interaction model increases the potential for genetic improvement. Including genomic information increases the prediction accuracy of estimated breeding values and reduces their bias and dispersion, although some biases persist. The results of the genome-wide association study indicate that CSD has a polygenic architecture and no major quantitative trait locus was detected.

First large-scale genomic prediction in the honey bee

Genomic selection has increased genetic gain in several livestock species, but due to the complicated genetics and reproduction biology not yet in honey bees. Recently, 2970 queens were genotyped to gather a reference population. For the application of genomic selection in honey bees, this study analyzes the accuracy and bias of pedigree-based and genomic breeding values for honey yield, three workability traits, and two traits for resistance against the parasite Varroa destructor. For breeding value estimation, we use a honey bee-specific model with maternal and direct effects, to account for the contributions of the workers and the queen of a colony to the phenotypes. We conducted a validation for the last generation and a five-fold cross-validation. In the validation for the last generation, the accuracy of pedigree-based estimated breeding values was 0.12 for honey yield, and ranged from 0.42 to 0.61 for the workability traits. The inclusion of genomic marker data improved these accuracies to 0.23 for honey yield, and a range from 0.44 to 0.65 for the workability traits. The inclusion of genomic data did not improve the accuracy of the disease-related traits. Traits with high heritability for maternal effects compared to the heritability for direct effects showed the most promising results. For all traits except the Varroa resistance traits, the bias with genomic methods was on a similar level compared to the bias with pedigree-based BLUP. The results show that genomic selection can successfully be applied to honey bees.

Low-density SNP panel for efficient imputation and genomic selection of milk production and technological traits in dairy sheep

The present study aimed to ascertain how different strategies for leveraging genomic information enhance the accuracy of estimated breeding values for milk and cheese-making traits and to evaluate the implementation of a low-density (LowD) SNP chip designed explicitly for that aim. Thus, milk samples from a total of 2,020 dairy ewes from 2 breeds (1,039 Spanish Assaf and 981 Churra) were collected and analyzed to determine 3 milk production and composition traits and 2 traits related to milk coagulation properties and cheese yield. The 2 studied populations were genotyped with a customized 50K Affymetrix SNP chip (Affymetrix Inc.) containing 55,627 SNP markers. The prediction accuracies were obtained using different multitrait methodologies, such as the BLUP model based on pedigree information, the genomic BLUP (GBLUP), and the BLUP at the SNP level (SNP-BLUP), which are based on genotypic data, and the single-step GBLUP (ssGBLUP), which combines both sources of information. All of these methods were analyzed by cross-validation, comparing predictions of the whole population with the test population sets. Additionally, we describe the design of a LowD SNP chip (3K) and its prediction accuracies through the different methods mentioned previously. Furthermore, the results obtained using the LowD SNP chip were compared with those based on the 50K SNP chip data sets. Finally, we conclude that implementing genomic selection through the ssGBLUP model in the current breeding programs would increase the accuracy of the estimated breeding values compared with the BLUP methodology in the Assaf (from 0.19 to 0.39) and Churra (from 0.27 to 0.44) dairy sheep populations. The LowD SNP chip is cost-effective and has proven to be an accurate tool for estimating genomic breeding values for milk and cheese-making traits, microsatellite imputation, and parentage verification. The results presented here suggest that the routine use of this LowD SNP chip could potentially increase the genetic gains of the breeding selection programs of the 2 Spanish dairy sheep breeds considered here.

Accuracy of genomic selection for reducing susceptibility to pendulous crop in turkey (Meleagris gallopavo)

Pendulous crop (PC) in the turkey occurs when the crop distends from its normal position, thereby preventing the movement of feed and water from the crop down into the digestive system. This condition negatively impacts the turkey industry at both production and welfare levels. In this study, we estimated the genetic parameters for PC incidence and its genetic correlation with 5 production traits. Additionally, we evaluated the prediction accuracy and bias of breeding values for the selection candidates using pedigree (BLUP) or pedigree-genomic (ssGBLUP) relationships among the animals. A total of 245,783 turkey records were made available by Hybrid Turkeys, Kitchener, Canada. Of these, 6,545 were affected with PC. In addition, the data included 9,634 records for breast meat yield (BMY); 5,592 records for feed conversion ratio (FCR) and residual feed intake (RFI) in males; 170,844 records for body weight (BW) and walking score (WS) between 18 and 20 wk of age for males (71,012) and females (99,832), respectively. Among this population, 36,830 were genotyped using a 65K SNP Illumina Inc. chip. While all animals passed the quality control criteria, only 53,455 SNP markers were retained for subsequent analysis. Heritability for PC was estimated at 0.16 ± 0.00 and 0.17 ± 0.00 using BLUP and ssGBLUP, respectively. The incidence of PC was not genetically correlated with WS or FCR. Low unfavourable genetic correlations with BW (0.12 and 0.14), BMY (0.24 and 0.24) and RFI (-0.33 and -0.28) were obtained using BLUP and ssGBLUP, respectively. Using ssGBLUP showed higher prediction accuracy (0.51) for the breeding values for the selection candidates than the pedigree-based model (0.35). Whereas the bias of the prediction was slightly reduced with ssGBLUP (0.33 ± 0.05) than BLUP (0.30 ± 0.08), both models showed a regression coefficient lower than one, indicating inflation in the predictions. The results of this study suggest that PC is a heritable trait and selection for lower PC incidence rates is feasible. Although further investigation is necessary, selection for BW, BMY, and RFI may increase PC incidence. Incorporating genomic information would lead to higher accuracy in predicting the genetic merit for selection candidates.

Single-step genomic BLUP enables joint analysis of disconnected breeding programs: an example with Eucalyptus globulus Labill

Single-step GBLUP (HBLUP) efficiently combines genomic, pedigree, and phenotypic information for holistic genetic analyses of disjunct breeding populations. We combined data from two independent multigenerational Eucalyptus globulus breeding populations to provide direct comparisons across the programs and indirect predictions in environments where pedigreed families had not been evaluated. Despite few known pedigree connections between the programs, genomic relationships provided the connectivity required to create a unified relationship matrix, H, which was used to compare pedigree-based and HBLUP models. Stem volume data from 48 sites spread across three regions of southern Australia and wood quality data across 20 sites provided comparisons of model accuracy. Genotyping proved valuable for correcting pedigree errors and HBLUP more precisely defines relationships within and among populations, with relationships among the genotyped individuals used to connect the pedigrees of the two programs. Cryptic relationships among the native range populations provided evidence of population structure and evidence of the origin of landrace populations. HBLUP across programs improved the prediction accuracy of parents and genotyped individuals and enabled breeding value predictions to be directly compared and inferred in regions where little to no testing has been undertaken. The impact of incorporating genetic groups in the estimation of H will further align traditional genetic evaluation pipelines with approaches that incorporate marker-derived relationships into prediction models.

Genetic analysis of egg production traits in turkeys (Meleagris gallopavo) using a single-step genomic random regression model

BACKGROUND

Egg production traits are economically important in poultry breeding programs. Previous studies have shown that incorporating genomic data can increase the accuracy of genetic prediction of egg production. Our objective was to estimate the genetic and phenotypic parameters of such traits and compare the prediction accuracy of pedigree-based random regression best linear unbiased prediction (RR-PBLUP) and genomic single-step random regression BLUP (RR-ssGBLUP). Egg production was recorded on 7422 birds during 24 consecutive weeks from first egg laid. Hatch-week of birth by week of lay and week of lay by age at first egg were fitted as fixed effects and body weight as a covariate, while additive genetic and permanent environment effects were fitted as random effects, along with heterogeneous residual variances over 24 weeks of egg production. Predictions accuracies were compared based on two statistics: (1) the correlation between estimated breeding values and phenotypes divided by the square root of the trait heritability, and (2) the ratio of the variance of BLUP predictions of individual Mendelian sampling effects divided by one half of the estimate of the additive genetic variance.

RESULTS

Heritability estimates along the production trajectory obtained with RR-PBLUP ranged from 0.09 to 0.22, with higher estimates for intermediate weeks. Estimates of phenotypic correlations between weekly egg production were lower than the corresponding genetic correlation estimates. Our results indicate that genetic correlations decreased over the laying period, with the highest estimate being between traits in later weeks and the lowest between early weeks and later ages. Prediction accuracies based on the correlation-based statistic ranged from 0.11 to 0.44 for RR-PBLUP and from 0.22 to 0.57 for RR-ssGBLUP using the correlation-based statistic. The ratios of the variances of BLUP predictions of Mendelian sampling effects and one half of the additive genetic variance ranged from 0.17 to 0.26 for RR-PBLUP and from 0.17 to 0.34 for RR-ssGBLUP. Although the improvement in accuracies from RR-ssGBLUP over those from RR-PBLUP was not uniform over time for either statistic, accuracies obtained with RR-ssGBLUP were generally equal to or higher than those with RR-PBLUP.

CONCLUSIONS

Our findings show the potential advantage of incorporating genomic data in genetic evaluation of egg production traits using random regression models, which can contribute to the genetic improvement of egg production in turkey populations.

Optimized grouping to increase accuracy of prediction of breeding values based on group records in genomic selection breeding programs

BACKGROUND

Phenotypic records of group means or group sums are a good alternative to individual records for some difficult to measure, but economically important traits such as feed efficiency or egg production. Accuracy of predicted breeding values based on group records increases with increasing relationships between group members. The classical way to form groups with more closely-related animals is based on pedigree information. When genotyping information is available before phenotyping, its use to form groups may further increase the accuracy of prediction from group records. This study analyzed two grouping methods based on genomic information: (1) unsupervised clustering implemented in the STRUCTURE software and (2) supervised clustering that models genomic relationships.

RESULTS

Using genomic best linear unbiased prediction (GBLUP) models, estimates of the genetic variance based on group records were consistent with those based on individual records. When genomic information was available to constitute the groups, genomic relationship coefficients between group members were higher than when random grouping of paternal half-sibs and of full-sibs was applied. Grouping methods that are based on genomic information resulted in higher accuracy of genomic estimated breeding values (GEBV) prediction compared to random grouping. The increase was ~ 1.5% for full-sibs and ~ 11.5% for paternal half-sibs. In addition, grouping methods that are based on genomic information led to lower coancestry coefficients between the top animals ranked by GEBV. Of the two proposed methods, supervised clustering was superior in terms of accuracy, computation requirements and applicability. By adding surplus genotyped offspring (more genotyped offspring than required to fill the groups), the advantage of supervised clustering increased by up to 4.5% compared to random grouping of full-sibs, and by 14.7% compared to random grouping of paternal half-sibs. This advantage also increased with increasing family sizes or decreasing genome sizes.

CONCLUSIONS

The use of genotyping information for grouping animals increases the accuracy of selection when phenotypic group records are used in genomic selection breeding programs.

Use of genomic information to exploit genotype-by-environment interactions for body weight of broiler chicken in bio-secure and production environments

BACKGROUND

The increase in accuracy of prediction by using genomic information has been well-documented. However, benefits of the use of genomic information and methodology for genetic evaluations are missing when genotype-by-environment interactions (G × E) exist between bio-secure breeding (B) environments and commercial production (C) environments. In this study, we explored (1) G × E interactions for broiler body weight (BW) at weeks 5 and 6, and (2) the benefits of using genomic information for prediction of BW traits when selection candidates were raised and tested in a B environment and close relatives were tested in a C environment.

METHODS

A pedigree-based best linear unbiased prediction (BLUP) multivariate model was used to estimate variance components and predict breeding values (EBV) of BW traits at weeks 5 and 6 measured in B and C environments. A single-step genomic BLUP (ssGBLUP) model that combined pedigree and genomic information was used to predict EBV. Cross-validations were based on correlation, mean difference and regression slope statistics for EBV that were estimated from full and reduced datasets. These statistics are indicators of population accuracy, bias and dispersion of prediction for EBV of traits measured in B and C environments. Validation animals were genotyped and non-genotyped birds in the B environment only.

RESULTS

Several indications of G × E interactions due to environmental differences were found for BW traits including significant re-ranking, heterogeneous variances and different heritabilities for BW measured in environments B and C. The genetic correlations between BW traits measured in environments B and C ranged from 0.48 to 0.54. The use of combined pedigree and genomic information increased population accuracy of EBV, and reduced bias of EBV prediction for genotyped birds compared to the use of pedigree information only. A slight increase in accuracy of EBV was also observed for non-genotyped birds, but the bias of EBV prediction increased for non-genotyped birds.

CONCLUSIONS

The G × E interaction was strong for BW traits of broilers measured in environments B and C. The use of combined pedigree and genomic information increased population accuracy of EBV substantially for genotyped birds in the B environment compared to the use of pedigree information only.

Application of single-step genomic evaluation using social genetic effect model for growth in pig

Objective

Social genetic effects (SGE) are an important genetic component for growth, group productivity, and welfare in pigs. The present study was conducted to evaluate i) the feasibility of the single-step genomic best linear unbiased prediction (ssGBLUP) approach with the inclusion of SGE in the model in pigs, and ii) the changes in the contribution of heritable SGE to the phenotypic variance with different scaling ω constants for genomic relationships.

Methods

The dataset included performance tested growth rate records (average daily gain) from 13,166 and 21,762 pigs Landrace (LR) and Yorkshire (YS), respectively. A total of 1,041 (LR) and 964 (YS) pigs were genotyped using the Illumina PorcineSNP60 v2 BeadChip panel. With the BLUPF90 software package, genetic parameters were estimated using a modified animal model for competitive traits. Giving a fixed weight to pedigree relationships (τ: 1), several weights (ω_xx, 0.1 to 1.0; with a 0.1 interval) were scaled with the genomic relationship for best model fit with Akaike information criterion (AIC).

Results

The genetic variances and total heritability estimates (T²) were mostly higher with ssGBLUP than in the pedigree-based analysis. The model AIC value increased with any level of ω other than 0.6 and 0.5 in LR and YS, respectively, indicating the worse fit of those models. The theoretical accuracies of direct and social breeding value were increased by decreasing ω in both breeds, indicating the better accuracy of ω_0.1 models. Therefore, the optimal values of ω to minimize AIC and to increase theoretical accuracy were 0.6 in LR and 0.5 in YS.

Conclusion

In conclusion, single-step ssGBLUP model fitting SGE showed significant improvement in accuracy compared with the pedigree-based analysis method; therefore, it could be implemented in a pig population for genomic selection based on SGE, especially in South Korean populations, with appropriate further adjustment of tuning parameters for relationship matrices.

Opportunities for genomic selection in American mink: A simulation study

Genomic selection can be considered as an effective tool for developing breeding programs in American mink. However, the genetic gains for economically important traits can be influenced by the accuracy of genomic predictions. The objective of this study was to investigate the prediction accuracies of traditional best linear unbiased prediction (BLUP), multi-step genomic BLUP (GBLUP) and single-step GBLUP (ssGBLUP) methods in American mink using simulated data with different levels of heritability, marker density, training set (TS) sizes and selection designs based on either phenotypic performance or estimated breeding values (EBVs). Under EBV selection design, the accuracy of BLUP predictions was increased by 38% and 44% for h2 = 0.10, 27% and 29% for h2 = 0.20, and 5.8% and 6% for h2 = 0.50 using GBLUP and ssGBLUP methods, respectively. Under phenotypic selection design, the accuracies of prediction by ssGBLUP method were 11.8% and 15.4% higher than those obtained by GBLUP for heritability of 0.10 and 0.20, respectively. However, the efficiency of ssGBLUP and GBLUP was not influenced by selection design at higher level of heritability (h2 = 0.50). Furthermore, higher selection intensity increased the bias of predictions in both pedigree-based and genomic evaluations. Regardless of selection design, TS sizes for GBLUP and ssGBLUP methods should be at least 3000 to achieve more accuracy than using BLUP for heritability of 0.50 and marker density of 10k and 50k. Overall, more accurate predictions were obtained using ssGBLUP method particularly for lowly heritable traits and low density of markers. Our results indicated that TS sizes should be optimized in accordance with heritability level, marker density, selection design and prediction method for genomic selection in American mink. The results provided an initial framework for designing genomic selection in mink breeding programs.

Semi-parametric estimates of population accuracy and bias of predictions of breeding values and future phenotypes using the LR method

Background

Cross-validation tools are used increasingly to validate and compare genetic evaluation methods but analytical properties of cross-validation methods are rarely described. There is also a lack of cross-validation tools for complex problems such as prediction of indirect effects (e.g. maternal effects) or for breeding schemes with small progeny group sizes.

Results

We derive the expected value of several quadratic forms by comparing genetic evaluations including “partial” and “whole” data. We propose statistics that compare genetic evaluations including “partial” and “whole” data based on differences in means, covariance, and correlation, and term the use of these statistics “method LR” (from linear regression). Contrary to common belief, the regression of true on estimated breeding values is (on expectation) lower than 1 for small or related validation sets, due to family structures. For validation sets that are sufficiently large, we show that these statistics yield estimators of bias, slope or dispersion, and population accuracy for estimated breeding values. Similar results hold for prediction of future phenotypes although we show that estimates of bias, slope or dispersion using prediction of future phenotypes are sensitive to incorrect heritabilities or precorrection for fixed effects. We present an example for a set of 2111 Brahman beef cattle for which, in repeated partitioning of the data into training and validation sets, there is very good agreement of statistics of method LR with prediction of future phenotypes.

Conclusions

Analytical properties of cross-validation measures are presented. We present a new method named LR for cross-validation that is automatic, easy to use, and which yields the quantities of interest. The method compares predictions based on partial and whole data, which results in estimates of accuracy and biases. Prediction of observed records may yield biased results due to precorrection or use of incorrect heritabilities.

Electronic supplementary material

The online version of this article (10.1186/s12711-018-0426-6) contains supplementary material, which is available to authorized users.

The genetic connectedness calculated from genomic information and its effect on the accuracy of genomic prediction

The magnitude of connectedness among management units (e.g., flocks and herds) gives a reliable estimate of genetic evaluation across these units. Traditionally, pedigree-based methods have been used to evaluate the genetic connectedness in China. However, these methods have not been able to yield a substantial outcome due to the lack of accuracy and integrity of pedigree data. Therefore, it is necessary to ascertain genetic connectedness using genomic information (i.e., genome-based genetic connectedness). Moreover, the effects of various levels of genome-based genetic connectedness on the accuracy of genomic prediction still remain poorly understood. A simulation study was performed to evaluate the genome-based genetic connectedness across herds by applying prediction error variance of difference (PEVD), coefficient of determination (CD) and prediction error correlation (r). Genomic estimated breeding values (GEBV) were predicted using a GBLUP model from a single and joint reference population. Overall, a continued increase in CD and r with a corresponding decrease in PEVD was observed as the number of common sires varies from 0 to 19 regardless of heritability levels, indicating increasing genetic connectedness between herds. Higher heritability tends to obtain stronger genetic connectedness. Compared to pedigree information, genomic relatedness inferred from genomic information increased the estimates of genetic connectedness across herds. Genomic prediction using the joint versus single reference population increased the accuracy of genomic prediction by 25% and lower heritability benefited more. Moreover, the largest benefits were observed as the number of common sires equals 0, and the gain of accuracy decreased as the number of common sires increased. We confirmed that genome-based genetic connectedness enhanced the estimates of genetic connectedness across management units. Additionally, using the combined reference population substantially increased accuracy of genomic prediction. However, care should be taken when combining reference data for closely related populations, which may give less reliable prediction results.