Accuracy Of Genomic Prediction Research Articles

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.

Bias and accuracy of body weight trait evaluations of an F2 chicken using single-step genomic best linear unbiased prediction (ssGBLUP)

The objectives of this study were (i) to compare the accuracy and bias of estimates of breeding values for body weight (BW) at 2–7 weeks of age using pedigree-based best linear unbiased prediction (BLUP) and single-step genomic BLUP (ssGBLUP) methods, and (ii) to determine the best level of minor allele frequencies (MAFs) for pre-selection of SNPs for genomic prediction (GP). Records of 488 F2 broiler chickens obtained from crossbreeding of fast-growing Arian chickens and slow-growing Iranian native chickens at 2–7 weeks of age were used. Samples were genotyped using Illumina Chicken 60K BeadChip. To investigate the effect of MAFs on the accuracy of prediction, 48 379 quality-controlled SNPs were grouped into five subgroups with MAF bins 0.05–0.1, 0.1–0.2, 0.2–0.3, 0.3–0.4, and 0.4–0.5. Our results confirmed the superiority of ssGBLUP compared to traditional BLUP methodology. The average accuracy of GP improved by 59.03%, 220.34%, 0.46%, 5.61%, 0.45%, and 2.73% using ssGBLUP compared to BLUP for BW at 2–7 weeks of age, respectively. Depending on the age group, using a subset of SNPs with a specific MAF bin compared to all SNPs resulted in a remarkable improvement of GP accuracy for the observed traits.

Genome‐wide association and genomic prediction for yield and component traits ofMiscanthus sacchariflorus

AbstractAccelerating biomass improvement is a major goal ofMiscanthusbreeding. The development and implementation of genomic‐enabled breeding tools, like marker‐assisted selection (MAS) and genomic selection, has the potential to improve the efficiency ofMiscanthusbreeding. The present study conducted genome‐wide association (GWA) and genomic prediction of biomass yield and 14 yield‐components traits inMiscanthus sacchariflorus. We evaluated a diversity panel with 590 accessions ofM. sacchariflorusgrown across 4 years in one subtropical and three temperate locations and genotyped with 268,109 single‐nucleotide polymorphisms (SNPs). The GWA study identified a total of 835 significant SNPs and 674 candidate genes across all traits and locations. Of the significant SNPs identified, 280 were localized in mapped quantitative trait loci intervals and proximal to SNPs identified for similar traits in previously reportedMiscanthusstudies, providing additional support for the importance of these genomic regions for biomass yield. Our study gave insights into the genetic basis for yield‐component traits inM. sacchariflorusthat may facilitate marker‐assisted breeding for biomass yield. Genomic prediction accuracy for the yield‐related traits ranged from 0.15 to 0.52 across all locations and genetic groups. Prediction accuracies within the six genetic groupings ofM. saccharifloruswere limited due to low sample sizes. Nevertheless, the Korea/NE China/Russia (N = 237) genetic group had the highest prediction accuracy of all genetic groups (ranging 0.26–0.71), suggesting that with adequate sample sizes, there is strong potential for genomic selection within the genetic groupings ofM. sacchariflorus. This study indicated that MAS and genomic prediction will likely be beneficial for conducting population‐improvement ofM. sacchariflorus.

Haplotype blocks for genomic prediction: a comparative evaluation in multiple crop datasets.

In modern plant breeding, genomic selection is becoming the gold standard for selection of superior genotypes. The basis for genomic prediction models is a set of phenotyped lines along with their genotypic profile. With high marker density and linkage disequilibrium (LD) between markers, genotype data in breeding populations tends to exhibit considerable redundancy. Therefore, interest is growing in the use of haplotype blocks to overcome redundancy by summarizing co-inherited features. Moreover, haplotype blocks can help to capture local epistasis caused by interacting loci. Here, we compared genomic prediction methods that either used single SNPs or haplotype blocks with regards to their prediction accuracy for important traits in crop datasets. We used four published datasets from canola, maize, wheat and soybean. Different approaches to construct haplotype blocks were compared, including blocks based on LD, physical distance, number of adjacent markers and the algorithms implemented in the software "Haploview" and "HaploBlocker". The tested prediction methods included Genomic Best Linear Unbiased Prediction (GBLUP), Extended GBLUP to account for additive by additive epistasis (EGBLUP), Bayesian LASSO and Reproducing Kernel Hilbert Space (RKHS) regression. We found improved prediction accuracy in some traits when using haplotype blocks compared to SNP-based predictions, however the magnitude of improvement was very trait- and model-specific. Especially in settings with low marker density, haplotype blocks can improve genomic prediction accuracy. In most cases, physically large haplotype blocks yielded a strong decrease in prediction accuracy. Especially when prediction accuracy varies greatly across different prediction models, prediction based on haplotype blocks can improve prediction accuracy of underperforming models. However, there is no "best" method to build haplotype blocks, since prediction accuracy varied considerably across methods and traits. Hence, criteria used to define haplotype blocks should not be viewed as fixed biological parameters, but rather as hyperparameters that need to be adjusted for every dataset.

Improved genomic prediction using machine learning with Variational Bayesian sparsity

BackgroundGenomic prediction has become a powerful modelling tool for assessing line performance in plant and livestock breeding programmes. Among the genomic prediction modelling approaches, linear based models have proven to provide accurate predictions even when the number of genetic markers exceeds the number of data samples. However, breeding programmes are now compiling data from large numbers of lines and test environments for analyses, rendering these approaches computationally prohibitive. Machine learning (ML) now offers a solution to this problem through the construction of fully connected deep learning architectures and high parallelisation of the predictive task. However, the fully connected nature of these architectures immediately generates an over-parameterisation of the network that needs addressing for efficient and accurate predictions.ResultsIn this research we explore the use of an ML architecture governed by variational Bayesian sparsity in its initial layers that we have called VBS-ML. The use of VBS-ML provides a mechanism for feature selection of important markers linked to the trait, immediately reducing the network over-parameterisation. Selected markers then propagate to the remaining fully connected feed-forward components of the ML network to form the final genomic prediction. We illustrated the approach with four large Australian wheat breeding data sets that range from 2665 lines to 10375 lines genotyped across a large set of markers. For all data sets, the use of the VBS-ML architecture improved genomic prediction accuracy over legacy linear based modelling approaches.ConclusionsAn ML architecture governed under a variational Bayesian paradigm was shown to improve genomic prediction accuracy over legacy modelling approaches. This VBS-ML approach can be used to dramatically decrease the parameter burden on the network and provide a computationally feasible approach for improving genomic prediction conducted with large breeding population numbers and genetic markers.

Alternative measures to evaluate the accuracy and bias of genomic predictions with censored records

This study aimed to propose and compare metrics of accuracy and bias of genomic prediction of breeding values for traits with censored data. Genotypic and censored-phenotypic information were simulated for four traits with QTL heritability and polygenic heritability, respectively: C1: 0.07-0.07, C2: 0.07-0.00, C3: 0.27-0.27, and C4: 0.27-0.00. Genomic breeding values were predicted using the Mixed Cox and Truncated Normal models. The accuracy of the models was estimated based on the Pearson (PC), maximal (MC), and Pearson correlation for censored data (PCC) while the genomic bias was calculated via simple linear regression (SLR) and Tobit (TB). MC and PCC were statistically superior to PC for the trait C3 with 10 and 40% censored information, for 70% censorship, PCC yielded better results than MC and PC. For the other traits, the proposed measures were superior or statistically equal to the PC. The coefficients associated with the marginal effects (TB) presented estimates close to those obtained for the SLR method, while the coefficient related to the latent variable showed almost unchanged pattern with the increase in censorship in most cases. From a statistical point of view, the use of methodologies for censored data should be prioritized, even for low censoring percentages.

Potential of low-density genotype imputation for cost-efficient genomic selection for resistance to Flavobacterium columnare in rainbow trout (Oncorhynchus mykiss)

BackgroundFlavobacterium columnare is the pathogen agent of columnaris disease, a major emerging disease that affects rainbow trout aquaculture. Selective breeding using genomic selection has potential to achieve cumulative improvement of the host resistance. However, genomic selection is expensive partly because of the cost of genotyping large numbers of animals using high-density single nucleotide polymorphism (SNP) arrays. The objective of this study was to assess the efficiency of genomic selection for resistance to F. columnare using in silico low-density (LD) panels combined with imputation. After a natural outbreak of columnaris disease, 2874 challenged fish and 469 fish from the parental generation (n = 81 parents) were genotyped with 27,907 SNPs. The efficiency of genomic prediction using LD panels was assessed for 10 panels of different densities, which were created in silico using two sampling methods, random and equally spaced. All LD panels were also imputed to the full 28K HD panel using the parental generation as the reference population, and genomic predictions were re-evaluated. The potential of prioritizing SNPs that are associated with resistance to F. columnare was also tested for the six lower-density panels.ResultsThe accuracies of both imputation and genomic predictions were similar with random and equally-spaced sampling of SNPs. Using LD panels of at least 3000 SNPs or lower-density panels (as low as 300 SNPs) combined with imputation resulted in accuracies that were comparable to those of the 28K HD panel and were 11% higher than the pedigree-based predictions.ConclusionsCompared to using the commercial HD panel, LD panels combined with imputation may provide a more affordable approach to genomic prediction of breeding values, which supports a more widespread adoption of genomic selection in aquaculture breeding programmes.

Alternative SNP weighting for multi-step and single-step genomic BLUP in the presence of causative variants.

The accuracy of genetic selection in dairy can be increased by the adoption of new technologies, such as the inclusion of sequence data. In simulation studies, assigning different weights to causative single-nucleotide polymorphism (SNP) markers led to better predictions depending on the genomic prediction method used. However, it is still not clear how the weights should be calculated. Our objective was to evaluate the accuracy of a multi-step method (GBLUP) and single-step GBLUP with simulated data using regular SNP, causatives variants (QTN) and the combination of both. Additionally, we compared the accuracies of all previous scenarios using alternatives for SNP weighting. The data were simulated assuming a single trait with a heritability of 0.3. The effective population size (Ne) was approximately 200. The pedigree contained 440,000 animals, and approximately 16,800 individuals were genotyped. A total of 49,974 SNP markers were evenly placed throughout the genome, and 100, 1000 and 2000 causative QTN were simulated. Both GBLUP and ssGBLUP were used in this study. We evaluated quadratic and nonlinear SNP weights in addition to the unweighted G. The inclusion of QTN to panels led to significant accuracy gains. Nonlinear A was demonstrated to be superior to quadratic weighting and unweighted approaches; however, results from Nonlinear A were dependent on the equation parameters. The unweighted approach was more suitable for less polygenic scenarios. Finally, SNP weighting might help elucidate trait architecture features based on changes in the accuracy of genomic prediction.

A first look at the ability to use genomic prediction for improving the ratooning ability of sugarcane.

The sugarcane ratooning ability (RA) is the most important target trait for breeders seeking to enhance the profitability of sugarcane production by reducing the planting cost. Understanding the genetics governing the RA could help breeders by identifying molecular markers that could be used for genomics-assisted breeding (GAB). A replicated field trial was conducted for three crop cycles (plant cane, first ratoon, and second ratoon) using 432 sugarcane clones and used for conducting genome-wide association and genomic prediction of five sugar and yield component traits of the RA. The RA traits for economic index (EI), stalk population (SP), stalk weight (SW), tonns of cane per hectare (TCH), and tonns of sucrose per hectare (TSH) were estimated from the yield and sugar data. A total of six putative quantitative trait loci and eight nonredundant single-nucleotide polymorphism (SNP) markers were associated with all five tested RA traits and appear to be unique. Seven putative candidate genes were colocated with significant SNPs associated with the five RA traits. The genomic prediction accuracies for those tested traits were moderate and ranged from 0.21 to 0.36. However, the models fitting fixed effects for the most significant associated markers for each respective trait did not give any advantages over the standard models without fixed effects. As a result of this study, more robust markers could be used in the future for clone selection in sugarcane, potentially helping resolve the genetic control of the RA in sugarcane.

Multi‐trait genomic predictions using GBLUP and Bayesian mixture prior model in beef cattle

AbstractMultiple trait genomic selection incorporating correlated traits can improve the predictive ability of low‐heritability traits. In this study, we evaluated genomic prediction accuracy using multi‐trait BayesCπ method (MT‐BayesCπ), which allows for a broader range of mixture priors for important traits in beef cattle. We compared the prediction performance of MT‐BayesCπ with single‐trait genomic best linear unbiased prediction (ST‐GBLUP), multi‐trait GBLUP (MT‐GBLUP), and single‐trait BayeCπ (ST‐BayesCπ) methods. We found that ribeye area (REA) and ribeye weight (REWT) showed high heritability, while slaughter weight (SWT) and carcass weight (CWT) displayed medium heritability, and slaughter rate (SR) and feedlot average daily gain (FDG) showed low heritability. Highly positive genetic correlations were observed between CWT and SWT (0.981) and SR and REWT (0.921). Notably, the MT‐BayesCπ method showed superior predictive abilities compared to other models. Using MT‐BayesCπ method, the accuracy increased from 0.272 to 0.694 for CWT compared to ST‐GBLUP and ST‐BayesCπ. MT‐GBLUP and ST‐BayesCπ showed similar prediction accuracies, while MT‐BayesCπ showed the least biased evaluations. Additionally, our results suggested that prediction accuracy of low‐heritability traits significantly increased when they were combined with traits with high genetic correlation in a multi‐trait prediction. Our study suggests that multi‐trait genomic predictions using GBLUP and Bayesian mixture prior models is feasible for genomic selection in beef cattle. Our findings indicate that MT‐BayesCπ outperforms other models (ST‐GBLUP, MT‐GBLUP and ST‐BayesCπ), especially for low‐heritability traits.

DeepGBLUP: joint deep learning networks and GBLUP framework for accurate genomic prediction of complex traits in Korean native cattle

BackgroundGenomic prediction has become widespread as a valuable tool to estimate genetic merit in animal and plant breeding. Here we develop a novel genomic prediction algorithm, called deepGBLUP, which integrates deep learning networks and a genomic best linear unbiased prediction (GBLUP) framework. The deep learning networks assign marker effects using locally-connected layers and subsequently use them to estimate an initial genomic value through fully-connected layers. The GBLUP framework estimates three genomic values (additive, dominance, and epistasis) by leveraging respective genetic relationship matrices. Finally, deepGBLUP predicts a final genomic value by summing all the estimated genomic values.ResultsWe compared the proposed deepGBLUP with the conventional GBLUP and Bayesian methods. Extensive experiments demonstrate that the proposed deepGBLUP yields state-of-the-art performance on Korean native cattle data across diverse traits, marker densities, and training sizes. In addition, they show that the proposed deepGBLUP can outperform the previous methods on simulated data across various heritabilities and quantitative trait loci (QTL) effects.ConclusionsWe introduced a novel genomic prediction algorithm, deepGBLUP, which successfully integrates deep learning networks and GBLUP framework. Through comprehensive evaluations on the Korean native cattle data and simulated data, deepGBLUP consistently achieved superior performance across various traits, marker densities, training sizes, heritabilities, and QTL effects. Therefore, deepGBLUP is an efficient method to estimate an accurate genomic value. The source code and manual for deepGBLUP are available at https://github.com/gywns6287/deepGBLUP.

Predicting plasticity of rosette growth and metabolic fluxes in Arabidopsis thaliana.

Plants can rapidly mitigate the effects of suboptimal growth environments by phenotypic plasticity of fitness-traits. While genetic variation for phenotypic plasticity offers the means for breeding climate-resilient crop lines, accurate genomic prediction models for plasticity of fitness-related traits are still lacking. Here, we employed condition- and accession-specific metabolic models for 67 Arabidopsis thaliana accessions to dissect and predict plasticity of rosette growth to changes in nitrogen availability. We showed that specific reactions in photorespiration, linking carbon and nitrogen metabolism, as well as key pathways of central carbon metabolism exhibited substantial genetic variation for flux plasticity. We also demonstrated that, in comparison with a genomic prediction model for fresh weight (FW), genomic prediction of growth plasticity improves the predictability of FW under low nitrogen by 58.9% and by additional 15.4% when further integrating data on plasticity of metabolic fluxes. Therefore, the combination of metabolic and statistical modeling provides a stepping stone in understanding the molecular mechanisms and improving the predictability of plasticity for fitness-related traits.

DeepSATA: A Deep Learning-Based Sequence Analyzer Incorporating the Transcription Factor Binding Affinity to Dissect the Effects of Non-Coding Genetic Variants.

Utilizing large-scale epigenomics data, deep learning tools can predict the regulatory activity of genomic sequences, annotate non-coding genetic variants, and uncover mechanisms behind complex traits. However, these tools primarily rely on human or mouse data for training, limiting their performance when applied to other species. Furthermore, the limited exploration of many species, particularly in the case of livestock, has led to a scarcity of comprehensive and high-quality epigenetic data, posing challenges in developing reliable deep learning models for decoding their non-coding genomes. The cross-species prediction of the regulatory genome can be achieved by leveraging publicly available data from extensively studied organisms and making use of the conserved DNA binding preferences of transcription factors within the same tissue. In this study, we introduced DeepSATA, a novel deep learning-based sequence analyzer that incorporates the transcription factor binding affinity for the cross-species prediction of chromatin accessibility. By applying DeepSATA to analyze the genomes of pigs, chickens, cattle, humans, and mice, we demonstrated its ability to improve the prediction accuracy of chromatin accessibility and achieve reliable cross-species predictions in animals. Additionally, we showcased its effectiveness in analyzing pig genetic variants associated with economic traits and in increasing the accuracy of genomic predictions. Overall, our study presents a valuable tool to explore the epigenomic landscape of various species and pinpoint regulatory deoxyribonucleic acid (DNA) variants associated with complex traits.

A novel method for genomic-enabled prediction of cultivars in new environments.

Genomic selection (GS) has gained global importance due to its potential to accelerate genetic progress and improve the efficiency of breeding programs. In this research we proposed a method to improve the prediction accuracy of tested lines in new (untested) environments. The new method trained the model with a modified response variable (a difference of response variables) that decreases the lack of a non-stationary distribution between the training and testing and improved the prediction accuracy. We compared the prediction accuracy of the conventional genomic best linear unbiased prediction (GBLUP) model (M1) including (or not) genotype × environment interaction (GE) (M1_GE; M1_NO_GE) versus the proposed method (M2) on several data sets. The gain in prediction accuracy of M2, versus M1_GE, M1_NO_GE in terms of Pearson´s correlation was of at least 4.3%, while in terms of percentage of top-yielding lines captured when was selected the 10% (Best10) and 20% (Best20) of lines was at least of 19.5%, while in terms of Normalized Root Mean Squared Error (NRMSE) was of at least of 42.29%.

Do feature selection methods for selecting environmental covariables enhance genomic prediction accuracy?

Genomic selection (GS) is transforming plant and animal breeding, but its practical implementation for complex traits and multi-environmental trials remains challenging. To address this issue, this study investigates the integration of environmental information with genotypic information in GS. The study proposes the use of two feature selection methods (Pearson's correlation and Boruta) for the integration of environmental information. Results indicate that the simple incorporation of environmental covariates may increase or decrease prediction accuracy depending on the case. However, optimal incorporation of environmental covariates using feature selection significantly improves prediction accuracy in four out of six datasets between 14.25% and 218.71% under a leave one environment out cross validation scenario in terms of Normalized Root Mean Squared Error, but not relevant gain was observed in terms of Pearson´s correlation. In two datasets where environmental covariates are unrelated to the response variable, feature selection is unable to enhance prediction accuracy. Therefore, the study provides empirical evidence supporting the use of feature selection to improve the prediction power of GS.

Estimation of genetic correlations and genomic prediction accuracy for reproductive and carcass traits in Hanwoo cows

Estimation of genetic correlations and genomic prediction accuracy for reproductive and carcass traits in Hanwoo cows

Epistasis-aware genome-wide association studies provide insights into the efficient breeding of high-yield and high-quality rice

Marker-assisted selection (MAS) and genomic selection (GS) breeding have greatly improved the efficiency of rice breeding. Due to the influences of epistasis and gene pleiotropy, ensuring the actual breeding effect of MAS and GS is still a difficult challenge to overcome. In this study, 113 indica rice varieties (V) and their 565 testcross hybrids (TC) were used as the materials to investigate the genetic basis of 12 quality traits and nine agronomic traits. The original traits and general combining ability of the parents, as well as the original traits and mid-parent heterosis of TC, were subjected to genome-wide association analysis. In total, 381 primary significantly associated loci (SAL) and 1,759 secondary SALs that had epistatic interactions with these primary SALs were detected. Among these loci, 322 candidate genes located within or nearby the SALs were screened, 204 of which were cloned genes. A total of 39 MAS molecular modules that are beneficial for trait improvement were identified by pyramiding the superior haplotypes of candidate genes and desirable epistatic alleles of the secondary SALs. All the SALs were used to construct genetic networks, in which 91 pleiotropic loci were investigated. Additionally, we estimated the accuracy of genomic prediction in the parent V and TC by incorporating either no SALs, primary SALs, secondary SALs or epistatic effect SALs as covariates. Although the prediction accuracies of the four models were generally not significantly different in the TC dataset, the incorporation of primary SALs, secondary SALs, and epistatic effect SALs significantly improved the prediction accuracies of 5 (26%), 3 (16%), and 11 (58%) traits in the V dataset, respectively. These results suggested that SALs and epistatic effect SALs identified based on an additive genotype can provide considerable predictive power for the parental lines. They also provide insights into the genetic basis of complex traits and valuable information for molecular breeding in rice.

Use of genomic prediction to screen sorghum B-lines in hybrid testcrosses.

Use of trifluoromethanesulfonamide (TFMSA), a male gametocide, increases the opportunities to identify promising B-lines because large quantities of F1 seed can be generated prior to the laborious task of B-line sterilization. Combining TFMSA technology with genomic selection could efficiently evaluate sorghum B-lines in hybrid combination to maximize the rates of genetic gain of the crop. This study used two recombinant inbred B-line populations, consisting of 217 lines, which were testcrossed to two R-lines to produce 434 hybrids. Each population of testcross hybrids were evaluated across five environments. Population-based genomic prediction models were assessed across environments using three different cross-validation (CV) schemes, each with 70% training and 30% validation sets. The validation schemes were as follows: CV1-hybrids chosen randomly for validation; CV2-B-lines were randomly chosen, and each chosen B-line had one of the two corresponding testcross hybrids randomly chosen for the validation; and CV3-B-lines were randomly chosen, and each chosen B-line had both corresponding testcross hybrids chosen for the validation. CV1 and CV2 presented the highest prediction accuracies; nonetheless, the prediction accuracies of the CV schemes were not statistically different in many environments. We determined that combining the B-line populations could improve prediction accuracies, and the genomic prediction models were able to effectively rank the poorest 70% of hybrids even when genomic prediction accuracies themselves were low. Results indicate that combining genomic prediction models and TFMSA technology can effectively aid breeders in predicting B-line hybrid performance in early generations prior to the laborious task of generating A/B-line pairs.

Using Bayesian threshold model and machine learning method to improve the accuracy of genomic prediction for ordered categorical traits in fish

Ordered categorical traits are commonly used in fish breeding programs as they are easier to obtain than continuous observations. However, most studies treat ordered categorical traits as linear traits and analyze them using linear models, which can lead to a serious reduction in prediction accuracy by violating the basic assumptions of linear models. The aim of this study was to evaluate the advantages of Bayesian threshold model and machine learning method in genomic prediction of ordered categorical traits in fish. The study was based on the analyses of simulated data and real data of Atlantic salmon. Ordinal categorical traits were simulated with varying numbers of categories (2, 3 and 4) and levels of heritabilities (0.1, 0.3 and 0.5). Linear and threshold models with BayesA and BayesCπ methods, as well as a machine learning method, support vector regression with default (SVRdef) and tuning (SVRtuning) hyperparameters were used to investigate their prediction abilities. The results showed that Bayesian threshold models yielded 2.1%, 2.6% and 2.9% higher prediction accuracies on average for 2-, 3- and 4-category traits, respectively, than Bayesian linear models. Furthermore, SVRtuning produced higher prediction accuracy compared with SVRdef and Bayesian threshold models in all scenarios. For real data, Bayesian threshold models yielded 1.2% higher prediction accuracy than Bayesian linear models, and SVRdef and SVRtuning yielded 3.3% and 6.6% higher prediction accuracies than Bayesian methods, respectively. In conclusion, the use of Bayesian threshold model and machine learning method was beneficial for genomic prediction of ordered categorical traits in fish.

Effect of subdivision of the Lacaune dairy sheep breed on the accuracy of genomic prediction

Genomic selection was deployed in Lacaune dairy breed in 2015. Lacaune population split in 1972 into 2 breeding companies with associated flocks, and there have been very few exchanges of animals between the subpopulations, leading to divergence of the 2 subpopulations. In spite of that, there is a joint genomic prediction. The objective of this study is to understand how this structuring affects prediction accuracy. We analyzed all the data available from Lacaune breeding program for milk yield: around 6 million phenotypes, 2 million animals in the pedigree and more than 29,000 genotyped animals, including 3,434 and 2,868 AI rams for each company. To consider missing pedigree, we set up genetic groups using the theory of metafounders. First, we studied the pedigree and genomic structures of the 2 subpopulations calculating Fst, evolution of average pedigree relationships across time and principal components analysis of genomic relationships. In a second part, we compared the reliability between different scenarios: an evaluation with a single reference population (Alone), an evaluation with a joint reference population (Together) and an evaluation of one subpopulation based on the reference population of the other group (Indirect). The low Fst value (0.02) reveals that the 2 subpopulations are still genetically close. Nevertheless, a low and constant average relationship between the animals of the 2 subpopulations confirms the absence of recent connections between them. We can see with principal component analysis results that even if they are close, they diverge over time. Finally, we observe small gains in accuracy of Together versus Alone, in spite of whereas doubling the reference population size in Together. These gains vary across years and subpopulations: less than 0.08 (0.46 to 0.54; ratio of accuracy for the partial and whole evaluations-corresponding to the greatest change in this ratio for breeding company 1, observed for the cohort 2016) for one subpopulation and between 0.03 (0.55 to 0.58) and 0.17 (0.48 to 0.65) for the other. To conclude, the 2 subpopulations remain close enough genetically so that their combined evaluation is advantageous, even if only slightly.