Practical application of genomic selection in a doubled-haploid winter wheat breeding program

Crop improvement is a long-term, expensive institutional endeavor. Genomic selection (GS), which uses single nucleotide polymorphism (SNP) information to estimate genomic breeding values, has proven efficient to increasing genetic gain by accelerating the breeding process in animal breeding programs. As for crop improvement, with few exceptions, GS applicability remains in the evaluation of algorithm performance. In this study, we examined factors related to GS applicability in line development stage for grain yield using a hard red winter wheat (Triticum aestivum L.) doubled-haploid population. The performance of GS was evaluated in two consecutive years to predict grain yield. In general, the semi-parametric reproducing kernel Hilbert space prediction algorithm outperformed parametric genomic best linear unbiased prediction. For both parametric and semi-parametric algorithms, an upward bias in predictability was apparent in within-year cross-validation, suggesting the prerequisite of cross-year validation for a more reliable prediction. Adjusting the training population’s phenotype for genotype by environment effect had a positive impact on GS model’s predictive ability. Possibly due to marker redundancy, a selected subset of SNPs at an absolute pairwise correlation coefficient threshold value of 0.4 produced comparable results and reduced the computational burden of considering the full SNP set. Finally, in the context of an ongoing breeding and selection effort, the present study has provided a measure of confidence based on the deviation of line selection from GS results, supporting the implementation of GS in wheat variety development.     

Genomic Selection and Association Mapping in Rice (Oryza sativa): Effect of Trait Genetic Architecture,Training Population Composition,Marker Number and Statistical Model on Accuracy of Rice Genomic Selection in Elite,Tropical Rice Breeding Lines

Genomic Selection (GS) is a new breeding method in which genome-wide markers are used to predict the breeding value of individuals in a breeding population. GS has been shown to improve breeding efficiency in dairy cattle and several crop plant species, and here we evaluate for the first time its efficacy for breeding inbred lines of rice. We performed a genome-wide association study (GWAS) in conjunction with five-fold GS cross-validation on a population of 363 elite breeding lines from the International Rice Research Institute''s (IRRI) irrigated rice breeding program and herein report the GS results. The population was genotyped with 73,147 markers using genotyping-by-sequencing. The training population, statistical method used to build the GS model, number of markers, and trait were varied to determine their effect on prediction accuracy. For all three traits, genomic prediction models outperformed prediction based on pedigree records alone. Prediction accuracies ranged from 0.31 and 0.34 for grain yield and plant height to 0.63 for flowering time. Analyses using subsets of the full marker set suggest that using one marker every 0.2 cM is sufficient for genomic selection in this collection of rice breeding materials. RR-BLUP was the best performing statistical method for grain yield where no large effect QTL were detected by GWAS, while for flowering time, where a single very large effect QTL was detected, the non-GS multiple linear regression method outperformed GS models. For plant height, in which four mid-sized QTL were identified by GWAS, random forest produced the most consistently accurate GS models. Our results suggest that GS, informed by GWAS interpretations of genetic architecture and population structure, could become an effective tool for increasing the efficiency of rice breeding as the costs of genotyping continue to decline.     

Accuracy of Genomic Selection in a Rice Synthetic Population Developed for Recurrent Selection Breeding

Genomic selection (GS) is a promising strategy for enhancing genetic gain. We investigated the accuracy of genomic estimated breeding values (GEBV) in four inter-related synthetic populations that underwent several cycles of recurrent selection in an upland rice-breeding program. A total of 343 S_2:4 lines extracted from those populations were phenotyped for flowering time, plant height, grain yield and panicle weight, and genotyped with an average density of one marker per 44.8 kb. The relative effect of the linkage disequilibrium (LD) and minor allele frequency (MAF) thresholds for selecting markers, the relative size of the training population (TP) and of the validation population (VP), the selected trait and the genomic prediction models (frequentist and Bayesian) on the accuracy of GEBVs was investigated in 540 cross validation experiments with 100 replicates. The effect of kinship between the training and validation populations was tested in an additional set of 840 cross validation experiments with a single genomic prediction model. LD was high (average r² = 0.59 at 25 kb) and decreased slowly, distribution of allele frequencies at individual loci was markedly skewed toward unbalanced frequencies (MAF average value 15.2% and median 9.6%), and differentiation between the four synthetic populations was low (F_ST ≤0.06). The accuracy of GEBV across all cross validation experiments ranged from 0.12 to 0.54 with an average of 0.30. Significant differences in accuracy were observed among the different levels of each factor investigated. Phenotypic traits had the biggest effect, and the size of the incidence matrix had the smallest. Significant first degree interaction was observed for GEBV accuracy between traits and all the other factors studied, and between prediction models and LD, MAF and composition of the TP. The potential of GS to accelerate genetic gain and breeding options to increase the accuracy of predictions are discussed.     

Genome-Wide Prediction Methods in Highly Diverse and Heterozygous Species: Proof-of-Concept through Simulation in Grapevine

Nowadays, genome-wide association studies (GWAS) and genomic selection (GS) methods which use genome-wide marker data for phenotype prediction are of much potential interest in plant breeding. However, to our knowledge, no studies have been performed yet on the predictive ability of these methods for structured traits when using training populations with high levels of genetic diversity. Such an example of a highly heterozygous, perennial species is grapevine. The present study compares the accuracy of models based on GWAS or GS alone, or in combination, for predicting simple or complex traits, linked or not with population structure. In order to explore the relevance of these methods in this context, we performed simulations using approx 90,000 SNPs on a population of 3,000 individuals structured into three groups and corresponding to published diversity grapevine data. To estimate the parameters of the prediction models, we defined four training populations of 1,000 individuals, corresponding to these three groups and a core collection. Finally, to estimate the accuracy of the models, we also simulated four breeding populations of 200 individuals. Although prediction accuracy was low when breeding populations were too distant from the training populations, high accuracy levels were obtained using the sole core-collection as training population. The highest prediction accuracy was obtained (up to 0.9) using the combined GWAS-GS model. We thus recommend using the combined prediction model and a core-collection as training population for grapevine breeding or for other important economic crops with the same characteristics.     

Evaluation and retrospective optimization of genomic selection for yield and disease resistance in spring barley

The general applicability of genomic selection (GS) to plant breeding and principles guiding its use have been established by simulation and empirical cross-validation studies. More recently, studies have demonstrated genetic gains over multiple cycles of selection in a variety of crop species. In this study, we provide additional evidence for the effectiveness of GS in an actual breeding program by demonstrating significant gains of 186.1 kg ha^?1 and ??1.85 ppm for grain yield and deoxynivalenol, respectively, two unfavorably correlated quantitative traits, across 3 cycles of selection in a spring six-row barley breeding population. With its general effectiveness established, the next step is to increase the accuracy of predictions used in GS and thereby increase genetic gains. For this, we first showed that updating the training population (TP) with phenotyped lines from recent breeding cycles, specifically selected lines, had an overall positive effect on prediction accuracy. Additionally, we investigated four recently proposed algorithms that seek to optimize the composition of a TP. Overall, the optimization algorithms improved prediction accuracy when compared to a randomly selected TP subset of the same size, but which algorithm performed best was dependent on the trait being predicted and other factors discussed within. This retrospective investigation highlights the importance of maintaining and optimizing the TP when using GS in applied breeding to maximize prediction accuracy, thereby maximizing gain from selection and resource utilization efficiency.     

Accuracy of genomic selection models in a large population of open-pollinated families in white spruce

Genomic selection (GS) is of interest in breeding because of its potential for predicting the genetic value of individuals and increasing genetic gains per unit of time. To date, very few studies have reported empirical results of GS potential in the context of large population sizes and long breeding cycles such as for boreal trees. In this study, we assessed the effectiveness of marker-aided selection in an undomesticated white spruce (Picea glauca (Moench) Voss) population of large effective size using a GS approach. A discovery population of 1694 trees representative of 214 open-pollinated families from 43 natural populations was phenotyped for 12 wood and growth traits and genotyped for 6385 single-nucleotide polymorphisms (SNPs) mined in 2660 gene sequences. GS models were built to predict estimated breeding values using all the available SNPs or SNP subsets of the largest absolute effects, and they were validated using various cross-validation schemes. The accuracy of genomic estimated breeding values (GEBVs) varied from 0.327 to 0.435 when the training and the validation data sets shared half-sibs that were on average 90% of the accuracies achieved through traditionally estimated breeding values. The trend was also the same for validation across sites. As expected, the accuracy of GEBVs obtained after cross-validation with individuals of unknown relatedness was lower with about half of the accuracy achieved when half-sibs were present. We showed that with the marker densities used in the current study, predictions with low to moderate accuracy could be obtained within a large undomesticated population of related individuals, potentially resulting in larger gains per unit of time with GS than with the traditional approach.     

Genomic predictions of breeding values in a cloned <Emphasis Type="Italic">Eucalyptus globulus</Emphasis> population in Chile

In Chile, an intensive Eucalyptus globulus clonal selection program is being carried out to increase forest productivity for pulp production. A breeding population was used to investigate the predicted ability of single nucleotide polymorphism (SNP) markers for genomic selection (GS). A total of 310 clones from 53 families were used. Stem volume and wood density were measured on all clones. Trees were genotyped at 12 K polymorphic markers using the EUChip60K genotype array. Genomic best linear unbiased prediction, Bayesian lasso regression, Bayes B, and Bayes C models were used to predict genomic estimated breeding values (GEBV). For cross-validation, 260 individuals were sampled for model training and 50 individuals for model validation, using 2 folds and 10 replications each. The average predictive ability estimates for wood density and stem volume across the models were 0.58 and 0.75, respectively. The average rank correlations were 0.59 and 0.71, respectively. Models produced very similar bias for both traits. When clones were ranked based on their GEBV, models had similar phenotypic mean for the top 10% of the clones. The predicted ability of markers will likely decrease if the models are used to predict GEBV of new material coming from the breeding program, because of a different marker–trait phase introduced by recombination. The results should be validated with larger populations and across two generations before routine applications of GS in E. globulus. We suggest that GS is a viable strategy to accelerate clonal selection program of E. globulus in Chile.     

Genomic prediction in CIMMYT maize and wheat breeding programs

Genomic selection (GS) has been implemented in animal and plant species, and is regarded as a useful tool for accelerating genetic gains. Varying levels of genomic prediction accuracy have been obtained in plants, depending on the prediction problem assessed and on several other factors, such as trait heritability, the relationship between the individuals to be predicted and those used to train the models for prediction, number of markers, sample size and genotype × environment interaction (GE). The main objective of this article is to describe the results of genomic prediction in International Maize and Wheat Improvement Center''s (CIMMYT''s) maize and wheat breeding programs, from the initial assessment of the predictive ability of different models using pedigree and marker information to the present, when methods for implementing GS in practical global maize and wheat breeding programs are being studied and investigated. Results show that pedigree (population structure) accounts for a sizeable proportion of the prediction accuracy when a global population is the prediction problem to be assessed. However, when the prediction uses unrelated populations to train the prediction equations, prediction accuracy becomes negligible. When genomic prediction includes modeling GE, an increase in prediction accuracy can be achieved by borrowing information from correlated environments. Several questions on how to incorporate GS into CIMMYT''s maize and wheat programs remain unanswered and subject to further investigation, for example, prediction within and between related bi-parental crosses. Further research on the quantification of breeding value components for GS in plant breeding populations is required.     

Development of a high-throughput SNP array for sea cucumber (Apostichopus japonicus) and its application in genomic selection with MCP regularized deep neural networks

High-throughput single nucleotide polymorphism (SNP) genotyping assays are powerful tools for genetic studies and genomic breeding applications for many species. Though large numbers of SNPs have been identified in sea cucumber (Apostichopus japonicus), but, as yet, no high-throughput genotyping platform is available for this species. In this study, we designed and developed a high-throughput 24 K SNP genotyping array named HaishenSNP24K for A. japonicus, based on the multi-objective-local optimization (MOLO) algorithm and HD-Marker genotyping method. The SNP array exhibited a relatively high genotyping call rate (> 96%), genotyping accuracy (>95%) and exhibited highly polymorphic in sea cucumber populations. In addition, we also assessed its application in genomic selection (GS). Deep neural networks (DNN) that can capture the complicated interactions of genes have been proposed as a promising tool in GS for SNP-based genomic prediction of complex traits in animal breeding. To overcome the problem of over-fitting when using the HaishenSNP24K array as high-dimensional DNN input, we developed minmax concave penalty (MCP) regularization for sparse deep neural networks (DNN-MCP) that finds an optimal sparse structure of a DNN by minimizing the square error subject to the non-convex penalty MCP on the parameters (weights and biases). Compared to two linear models, namely RR-GBLUP and Bayes B, and the nonlinear model DNN, DNN-MCP has greatly improved the genomic prediction ability for three quantitative traits (e.g., wet weight, dry weight and survival time) in the sea cucumber population. To the best of our knowledge, this is the first work to develop a high-throughput SNP array for A. japonicus and a new model DNN-MCP for genomic prediction of complex traits in GS. The present results provide evidence that supports the HaishenSNP24K array with DNN-MCP will be valuable for genetic studies and molecular breeding in A. japonicus.     

Efficiency of genomic selection with models including dominance effect in the context of Eucalyptus breeding

We developed a simulation study to test the efficiency of genomic selection (GS) in the case of Eucalyptus breeding. We simulated a recurrent selection scheme for clone production over four breeding cycles. Scenarios crossing broad sense heritabilities (H ²?=?0.6 and 0.1) and dominance to additive variance ratios (R?=?0.1; 0.5; and 1) were compared. GS was performed with 1,000 SNPs and 22 QTLs per Morgan and tested against phenotypic selection (PS) based on best linear unbiased prediction of parents and clones. When the training population was made up of the first cycle progeny tests and the candidate populations were the progeny tests of three successive cycles, GS accuracy decreased with breeding cycles (e.g., from 0.9 to 0.4 with H ²?=?0.6 and R?=?0.1), whereas PS presented constant performances (accuracy of 0.8 with H ²?=?0.6 and R?=?0.1). When the training population set was updated by associating data of previous cycles, GS accuracy was improved from 25 % to 418 %, especially with H ²?=?0.1. The GS model including dominance effects performed better in clone selection (genotypic value) when dominance effects were preponderant (R?=?1), heritability was high (H ²?=?0.6 and with an updated training set), but no improvement was detected for parent selection (breeding value). The genetic gains over cycles were lower with the GS method without updating the data set but, with an updated training set, were similar to PS. However, the genetic gain per unit time with GS was 1.5 to 3 times higher than with PS for breeding and clone populations. These results highlight the value of GS in Eucalyptus breeding.     

Optimizing whole-genomic prediction for autotetraploid blueberry breeding

Blueberry (Vaccinium spp.) is an important autopolyploid crop with significant benefits for human health. Apart from its genetic complexity, the feasibility of genomic prediction has been proven for blueberry, enabling a reduction in the breeding cycle time and increasing genetic gain. However, as for other polyploid crops, sequencing costs still hinder the implementation of genome-based breeding methods for blueberry. This motivated us to evaluate the effect of training population sizes and composition, as well as the impact of marker density and sequencing depth on phenotype prediction for the species. For this, data from a large real breeding population of 1804 individuals were used. Genotypic data from 86,930 markers and three traits with different genetic architecture (fruit firmness, fruit weight, and total yield) were evaluated. Herein, we suggested that marker density, sequencing depth, and training population size can be substantially reduced with no significant impact on model accuracy. Our results can help guide decisions toward resource allocation (e.g., genotyping and phenotyping) in order to maximize prediction accuracy. These findings have the potential to allow for a faster and more accurate release of varieties with a substantial reduction of resources for the application of genomic prediction in blueberry. We anticipate that the benefits and pipeline described in our study can be applied to optimize genomic prediction for other diploid and polyploid species.     

Genome-enabled prediction of genetic values using radial basis function neural networks

The availability of high density panels of molecular markers has prompted the adoption of genomic selection (GS) methods in animal and plant breeding. In GS, parametric, semi-parametric and non-parametric regressions models are used for predicting quantitative traits. This article shows how to use neural networks with radial basis functions (RBFs) for prediction with dense molecular markers. We illustrate the use of the linear Bayesian LASSO regression model and of two non-linear regression models, reproducing kernel Hilbert spaces (RKHS) regression and radial basis function neural networks (RBFNN) on simulated data and real maize lines genotyped with 55,000 markers and evaluated for several trait-environment combinations. The empirical results of this study indicated that the three models showed similar overall prediction accuracy, with a slight and consistent superiority of RKHS and RBFNN over the additive Bayesian LASSO model. Results from the simulated data indicate that RKHS and RBFNN models captured epistatic effects; however, adding non-signal (redundant) predictors (interaction between markers) can adversely affect the predictive accuracy of the non-linear regression models.     

Genomic prediction ability for yield-related traits in German winter barley elite material

Key message

Genomic prediction was evaluated in German winter barley breeding lines. In this material, prediction ability is strongly influenced by population structure and main determinant of prediction ability is the close genetic relatedness of the breeding material.

Abstract

To ensure breeding progress under changing environmental conditions the implementation and evaluation of new breeding methods is of crucial importance. Modern breeding approaches like genomic selection may significantly accelerate breeding progress. We assessed the potential of genomic prediction in a training population of 750 genotypes, consisting of multiple six-rowed winter barley (Hordeum vulgare L.) elite material families and old cultivars, which reflect the breeding history of barley in Germany. Crosses of parents selected from the training set were used to create a set of double-haploid families consisting of 750 genotypes. Those were used to confirm prediction ability estimates based on a cross-validation with the training set material using 11 different genomic prediction models. Population structure was inferred with dimensionality reduction methods like discriminant analysis of principle components and the influence of population structure on prediction ability was investigated. In addition to the size of the training set, marker density is of crucial importance for genomic prediction. We used genome-wide linkage disequilibrium and persistence of linkage phase as indicators to estimate that 11,203 evenly spaced markers are required to capture all QTL effects. Although a 9k SNP array does not contain a sufficient number of polymorphic markers for long-term genomic selection, we obtained fairly high prediction accuracies ranging from 0.31 to 0.71 for the traits earing, hectoliter weight, spikes per square meter, thousand kernel weight and yield and show that they result from the close genetic relatedness of the material. Our work contributes to designing long-term genetic prediction programs for barley breeding.

     

Island-Model Genomic Selection for Long-Term Genetic Improvement of Autogamous Crops

Acceleration of genetic improvement of autogamous crops such as wheat and rice is necessary to increase cereal production in response to the global food crisis. Population and pedigree methods of breeding, which are based on inbred line selection, are used commonly in the genetic improvement of autogamous crops. These methods, however, produce a few novel combinations of genes in a breeding population. Recurrent selection promotes recombination among genes and produces novel combinations of genes in a breeding population, but it requires inaccurate single-plant evaluation for selection. Genomic selection (GS), which can predict genetic potential of individuals based on their marker genotype, might have high reliability of single-plant evaluation and might be effective in recurrent selection. To evaluate the efficiency of recurrent selection with GS, we conducted simulations using real marker genotype data of rice cultivars. Additionally, we introduced the concept of an “island model” inspired by evolutionary algorithms that might be useful to maintain genetic variation through the breeding process. We conducted GS simulations using real marker genotype data of rice cultivars to evaluate the efficiency of recurrent selection and the island model in an autogamous species. Results demonstrated the importance of producing novel combinations of genes through recurrent selection. An initial population derived from admixture of multiple bi-parental crosses showed larger genetic gains than a population derived from a single bi-parental cross in whole cycles, suggesting the importance of genetic variation in an initial population. The island-model GS better maintained genetic improvement in later generations than the other GS methods, suggesting that the island-model GS can utilize genetic variation in breeding and can retain alleles with small effects in the breeding population. The island-model GS will become a new breeding method that enhances the potential of genomic selection in autogamous crops, especially bringing long-term improvement.     

Accuracy of genomic selection for a sib-evaluated trait using identity-by-state and identity-by-descent relationships

Background

GBLUP (genomic best linear unbiased prediction) uses high-density single nucleotide polymorphism (SNP) markers to construct genomic identity-by-state (IBS) relationship matrices. However, identity-by-descent (IBD) relationships can be accurately calculated for extremely sparse markers. Here, we compare the accuracy of prediction of genome-wide breeding values (GW-BV) for a sib-evaluated trait in a typical aquaculture population, assuming either IBS or IBD genomic relationship matrices, and by varying marker density and size of the training dataset.

Methods

A simulation study was performed, assuming a population with strong family structure over three subsequent generations. Traditional and genomic BLUP were used to estimate breeding values, the latter using either IBS or IBD genomic relationship matrices, with marker densities ranging from 10 to ~1200 SNPs/Morgan (M). Heritability ranged from 0.1 to 0.8, and phenotypes were recorded on 25 to 45 sibs per full-sib family (50 full-sib families). Models were compared based on their predictive ability (accuracy) with respect to true breeding values of unphenotyped (albeit genotyped) sibs in the last generation.

Results

As expected, genomic prediction had greater accuracy compared to pedigree-based prediction. At the highest marker density, genomic prediction based on IBS information (IBS-GS) was slightly superior to that based on IBD information (IBD-GS), while at lower densities (≤100 SNPs/M), IBD-GS was more accurate. At the lowest densities (10 to 20 SNPs/M), IBS-GS was even outperformed by the pedigree-based model. Accuracy of IBD-GS was stable across marker densities performing well even down to 10 SNPs/M (2.5 to 6.1% reduction in accuracy compared to ~1200 SNPs/M). Loss of accuracy due to reduction in the size of training datasets was moderate and similar for both genomic prediction models. The relative superiority of (high-density) IBS-GS over IBD-GS was more pronounced for traits with a low heritability.

Conclusions

Using dense markers, GBLUP based on either IBD or IBS relationship matrices proved to perform better than a pedigree-based model. However, accuracy of IBS-GS declined rapidly with decreasing marker densities, and was even outperformed by a traditional pedigree-based model at the lowest densities. In contrast, the accuracy of IBD-GS was very stable across marker densities.     

Identity-by-descent genomic selection using selective and sparse genotyping

Background

Genomic selection methods require dense and widespread genotyping data, posing a particular challenge if both sexes are subject to intense selection (e.g., aquaculture species). This study focuses on alternative low-cost genomic selection methods (IBD-GS) that use selective genotyping with sparse marker panels to estimate identity-by-descent relationships through linkage analysis. Our aim was to evaluate the potential of these methods in selection programs for continuous traits measured on sibs of selection candidates in a typical aquaculture breeding population.

Methods

Phenotypic and genomic data were generated by stochastic simulation, assuming low to moderate heritabilities (0.10 to 0.30) for a Gaussian trait measured on sibs of the selection candidates in a typical aquaculture breeding population that consisted of 100 families (100 training animals and 20 selection candidates per family). Low-density marker genotype data (~ 40 markers per Morgan) were used to trace genomic identity-by-descent relationships. Genotyping was restricted to selection candidates from 30 phenotypically top-ranking families and varying fractions of their phenotypically extreme training sibs. All phenotypes were included in the genetic analyses. Classical pedigree-based and IBD-GS models were compared based on realized genetic gain over one generation of selection.

Results

Genetic gain increased substantially (13 to 32%) with IBD-GS compared to classical selection and was greatest with higher heritability. Most of the extra gain from IBD-GS was obtained already by genotyping the 5% phenotypically most extreme sibs within the pre-selected families. Additional genotyping further increased genetic gains, but these were small when going from genotyping 20% of the extremes to all phenotyped sibs. The success of IBD-GS with sparse and selective genotyping can be explained by the fact that within-family haplotype blocks are accurately traced even with low-marker densities and that most of the within-family variance for normally distributed traits is captured by a small proportion of the phenotypically extreme sibs.

Conclusions

IBD-GS was substantially more effective than classical selection, even when based on very few markers and combined with selective genotyping of small fractions of the population. The study shows that low-cost GS programs can be successful by combining sparse and selective genotyping with pedigree and linkage information.     

Genomic selection in forest tree breeding

Genomic selection (GS) involves selection decisions based on genomic breeding values estimated as the sum of the effects of genome-wide markers capturing most quantitative trait loci (QTL) for the target trait(s). GS is revolutionizing breeding practice in domestic animals. The same approach and concepts can be readily applied to forest tree breeding where long generation times and late expressing complex traits are also a challenge. GS in forest trees would have additional advantages: large training populations can be easily assembled and accurately phenotyped for several traits, and the extent of linkage disequilibrium (LD) can be high in elite populations with small effective population size (N _e) frequently used in advanced forest tree breeding programs. Deterministic equations were used to assess the impact of LD (modeled by N _e and intermarker distance), the size of the training set, trait heritability, and the number of QTL on the predicted accuracy of GS. Results indicate that GS has the potential to radically improve the efficiency of tree breeding. The benchmark accuracy of conventional BLUP selection is reached by GS even at a marker density ~2 markers/cM when N _e ≤ 30, while up to 20 markers/cM are necessary for larger N _e. Shortening the breeding cycle by 50% with GS provides an increase ≥100% in selection efficiency. With the rapid technological advances and declining costs of genotyping, our cautiously optimistic outlook is that GS has great potential to accelerate tree breeding. However, further simulation studies and proof-of-concept experiments of GS are needed before recommending it for operational implementation.     

Towards genomic selection in apple (Malus ×domestica Borkh.) breeding programmes: Prospects, challenges and strategies

The apple genome sequence and the availability of high-throughput genotyping technologies have initiated a new era where SNP markers are abundant across the whole genome. Genomic selection (GS) is a statistical approach that utilizes all available genome-wide markers simultaneously to estimate breeding values or total genetic values. For breeding programmes, GS is a promising alternative to the traditional marker-assisted selection for manipulating complex polygenic traits often controlled by many small-effect genes. Various factors, such as genetic architecture of selection traits, population size and structure, genetic evaluation systems, density of SNP markers and extent of linkage disequilibrium, have been shown to be the key drivers of the accuracy of GS. In this paper, we provide an overview of the status of these aspects in current apple-breeding programmes. Strategies for GS for fruit quality and disease resistance are discussed, and an update on an empirical genomic selection study in a New Zealand apple-breeding programme is provided, along with a foresight of expected accuracy from such selection.     

Application of neural networks with back-propagation to genome-enabled prediction of complex traits in Holstein-Friesian and German Fleckvieh cattle

Background

Recently, artificial neural networks (ANN) have been proposed as promising machines for marker-based genomic predictions of complex traits in animal and plant breeding. ANN are universal approximators of complex functions, that can capture cryptic relationships between SNPs (single nucleotide polymorphisms) and phenotypic values without the need of explicitly defining a genetic model. This concept is attractive for high-dimensional and noisy data, especially when the genetic architecture of the trait is unknown. However, the properties of ANN for the prediction of future outcomes of genomic selection using real data are not well characterized and, due to high computational costs, using whole-genome marker sets is difficult. We examined different non-linear network architectures, as well as several genomic covariate structures as network inputs in order to assess their ability to predict milk traits in three dairy cattle data sets using large-scale SNP data. For training, a regularized back propagation algorithm was used. The average correlation between the observed and predicted phenotypes in a 20 times 5-fold cross-validation was used to assess predictive ability. A linear network model served as benchmark.

Results

Predictive abilities of different ANN models varied markedly, whereas differences between data sets were small. Dimension reduction methods enhanced prediction performance in all data sets, while at the same time computational cost decreased. For the Holstein-Friesian bull data set, an ANN with 10 neurons in the hidden layer achieved a predictive correlation of r=0.47 for milk yield when the entire marker matrix was used. Predictive ability increased when the genomic relationship matrix (r=0.64) was used as input and was best (r=0.67) when principal component scores of the marker genotypes were used. Similar results were found for the other traits in all data sets.

Conclusion

Artificial neural networks are powerful machines for non-linear genome-enabled predictions in animal breeding. However, to produce stable and high-quality outputs, variable selection methods are highly recommended, when the number of markers vastly exceeds sample size.     

Effect of genomic prediction on response to selection in forest tree breeding

Through stochastic simulations, estimates of breeding values accuracies and response to selection were assessed under traditional pedigree-based and genomic-based evaluation methods. More specifically, several key parameters such as the trait’s heritability (0.2 and 0.6), the number of QTLs underlying the trait (100 to 200), and the marker density (1 to 10 SNPs/cM) were evaluated. Additionally, impact of two contrasting mating designs (partial diallel vs. single-pair mating) was investigated. Response to selection was then assessed in a seed production population (seed orchard consisting of unrelated selections) for different effective population sizes (N_e?=?5 to 25). The simulated candidate population comprised a fixed size of 2050 individuals with fast linkage disequilibrium decay, generally found in forest tree populations. Following the genetic/genomic evaluation, top-ranked individuals were selected to meeting the predetermined effective population size in target production population. The combination of low h², high N_e, and dense marker coverage resulted at maximum relative genomic prediction efficiency and the most efficient exploitation of the Mendelian sampling term (within-family additive genetic variance). Since genomic prediction of breeding values constitutes the methodological foundation of genomic selection, our results can be used to address important questions when similar scenarios are considered.