Akio Onogi

Exploring the areas of applicability of whole-genome prediction methods for Asian rice (Oryza sativa L.)

Key messageOur simulation results clarify the areas of applicability of nine prediction methods and suggest the factors that affect their accuracy at predicting empirical traits.AbstractWhole-genome prediction is used to predict genetic value from genome-wide markers. The choice of method is important for successful prediction. We compared nine methods using empirical data for eight phenological and morphological traits of Asian rice cultivars (Oryza sativa L.) and data simulated from real marker genotype data. The methods were genomic BLUP (GBLUP), reproducing kernel Hilbert spaces regression (RKHS), Lasso, elastic net, random forest (RForest), Bayesian lasso (Blasso), extended Bayesian lasso (EBlasso), weighted Bayesian shrinkage regression (wBSR), and the average of all methods (Ave). The objectives were to evaluate the predictive ability of these methods in a cultivar population, to characterize them by exploring the area of applicability of each method using simulation, and to investigate the causes of their different accuracies for empirical traits. GBLUP was the most accurate for one trait, RKHS and Ave for two, and RForest for three traits. In the simulation, Blasso, EBlasso, and Ave showed stable performance across the simulated scenarios, whereas the other methods, except wBSR, had specific areas of applicability; wBSR performed poorly in most scenarios. For each method, the accuracy ranking for the empirical traits was largely consistent with that in one of the simulated scenarios, suggesting that the simulation conditions reflected the factors that affected the method accuracy for the empirical results. This study will be useful for genomic prediction not only in Asian rice, but also in populations from other crops with relatively small training sets and strong linkage disequilibrium structures.

Toward integration of genomic selection with crop modelling: the development of an integrated approach to predicting rice heading dates.

Key messageIt is suggested that accuracy in predicting plant phenotypes can be improved by integrating genomic prediction with crop modelling in a single hierarchical model.AbstractAccurate prediction of phenotypes is important for plant breeding and management. Although genomic prediction/selection aims to predict phenotypes on the basis of whole-genome marker information, it is often difficult to predict phenotypes of complex traits in diverse environments, because plant phenotypes are often influenced by genotype–environment interaction. A possible remedy is to integrate genomic prediction with crop/ecophysiological modelling, which enables us to predict plant phenotypes using environmental and management information. To this end, in the present study, we developed a novel method for integrating genomic prediction with phenological modelling of Asian rice (Oryza sativa, L.), allowing the heading date of untested genotypes in untested environments to be predicted. The method simultaneously infers the phenological model parameters and whole-genome marker effects on the parameters in a Bayesian framework. By cultivating backcross inbred lines of Koshihikari × Kasalath in nine environments, we evaluated the potential of the proposed method in comparison with conventional genomic prediction, phenological modelling, and two-step methods that applied genomic prediction to phenological model parameters inferred from Nelder–Mead or Markov chain Monte Carlo algorithms. In predicting heading dates of untested lines in untested environments, the proposed and two-step methods tended to provide more accurate predictions than the conventional genomic prediction methods, particularly in environments where phenotypes from environments similar to the target environment were unavailable for training genomic prediction. The proposed method showed greater accuracy in prediction than the two-step methods in all cross-validation schemes tested, suggesting the potential of the integrated approach in the prediction of phenotypes of plants.

A simulation-based breeding design that uses whole-genome prediction in tomato

Efficient plant breeding methods must be developed in order to increase yields and feed a growing world population, as well as to meet the demands of consumers with diverse preferences who require high-quality foods. We propose a strategy that integrates breeding simulations and phenotype prediction models using genomic information. The validity of this strategy was evaluated by the simultaneous genetic improvement of the yield and flavour of the tomato (Solanum lycopersicum), as an example. Reliable phenotype prediction models for the simulation were constructed from actual genotype and phenotype data. Our simulation predicted that selection for both yield and flavour would eventually result in morphological changes that would increase the total plant biomass and decrease the light extinction coefficient, an essential requirement for these improvements. This simulation-based genome-assisted approach to breeding will help to optimise plant breeding, not only in the tomato but also in other important agricultural crops.

Genome-wide association study and genomic prediction in citrus: Potential of genomics-assisted breeding for fruit quality traits

Novel genomics-based approaches such as genome-wide association studies (GWAS) and genomic selection (GS) are expected to be useful in fruit tree breeding, which requires much time from the cross to the release of a cultivar because of the long generation time. In this study, a citrus parental population (111 varieties) and a breeding population (676 individuals from 35 full-sib families) were genotyped for 1,841 single nucleotide polymorphisms (SNPs) and phenotyped for 17 fruit quality traits. GWAS power and prediction accuracy were increased by combining the parental and breeding populations. A multi-kernel model considering both additive and dominance effects improved prediction accuracy for acidity and juiciness, implying that the effects of both types are important for these traits. Genomic best linear unbiased prediction (GBLUP) with linear ridge kernel regression (RR) was more robust and accurate than GBLUP with non-linear Gaussian kernel regression (GAUSS) in the tails of the phenotypic distribution. The results of this study suggest that both GWAS and GS are effective for genetic improvement of citrus fruit traits. Furthermore, the data collected from breeding populations are beneficial for increasing the detection power of GWAS and the prediction accuracy of GS.

Genomic prediction in Japanese Black cattle: application of a single-step approach to beef cattle.

The implementation of genomic selection for Japanese Black cattle, known for rich marbling of their meat, is now being explored. Although multiple-step methods are often adopted for dairy cattle, they present shortcomings such as bias and loss of information in addition to operational complexity. These can be avoided using single-step genomic BLUP (ssGBLUP) based on the relationship matrix H, which is constructed from the numerator relationship matrix (A) augmented by the genomic relationship matrix (G). This study assessed the use of ssGBLUP for 3 economically important traits in Japanese Black cattle. Three aspects of ssGBLUP that are important for practical use were examined specifically: the mixing proportions of blending G with A, selection of subsets of genotyped animals used for constructing H, and prediction ability for ungenotyped animals. Different mixing proportions were tested to assess the influence of these proportions on variance component estimation and prediction accuracy. For all traits, the highest or nearly highest accuracy was obtained when the adopted mixing proportion provided heritability closest to that inferred based on A. However, the accuracy did not increase greatly under adjustment of the mixing proportion, thereby suggesting that the influence of the mixing proportion on the accuracy was limited. Genotype data of influential bulls showed a greater contribution to accuracy than that of bulls that were less influential. Genotyping animals with phenotypic records increased the accuracy. It can be prioritized over genotyping bulls that are not influential on the population. These results are expected to present good guides to the future expansion of genotyped populations. Even for animals without genotype data but with genotyped sires, ssGBLUP provided more accurate prediction than BLUP did. For both phenotype and breeding value prediction, ssGBLUP provides more accurate prediction than BLUP, suggesting its usefulness in genomic selection in Japanese Black cattle.

A Ranking Approach to Genomic Selection

Background Genomic selection (GS) is a recent selective breeding method which uses predictive models based on whole-genome molecular markers. Until now, existing studies formulated GS as the problem of modeling an individual’s breeding value for a particular trait of interest, i.e., as a regression problem. To assess predictive accuracy of the model, the Pearson correlation between observed and predicted trait values was used. Contributions In this paper, we propose to formulate GS as the problem of ranking individuals according to their breeding value. Our proposed framework allows us to employ machine learning methods for ranking which had previously not been considered in the GS literature. To assess ranking accuracy of a model, we introduce a new measure originating from the information retrieval literature called normalized discounted cumulative gain (NDCG). NDCG rewards more strongly models which assign a high rank to individuals with high breeding value. Therefore, NDCG reflects a prerequisite objective in selective breeding: accurate selection of individuals with high breeding value. Results We conducted a comparison of 10 existing regression methods and 3 new ranking methods on 6 datasets, consisting of 4 plant species and 25 traits. Our experimental results suggest that tree-based ensemble methods including McRank, Random Forests and Gradient Boosting Regression Trees achieve excellent ranking accuracy. RKHS regression and RankSVM also achieve good accuracy when used with an RBF kernel. Traditional regression methods such as Bayesian lasso, wBSR and BayesC were found less suitable for ranking. Pearson correlation was found to correlate poorly with NDCG. Our study suggests two important messages. First, ranking methods are a promising research direction in GS. Second, NDCG can be a useful evaluation measure for GS.

Whole-genome prediction of fatty acid composition in meat of Japanese Black cattle.

Because fatty acid composition influences the flavor and texture of meat, controlling it is particularly important for cattle breeds such as the Japanese Black, characterized by high meat quality. We evaluated the predictive ability of single-step genomic best linear unbiased prediction (ssGBLUP) in fatty acid composition of Japanese Black cattle by assessing the composition of seven fatty acids in 3088 cattle, of which 952 had genome-wide marker genotypes. All sires of the genotyped animals were genotyped, but their dams were not. Cross-validation was conducted for the 952 animals. The prediction accuracy was higher with ssGBLUP than with best linear unbiased prediction (BLUP) for all traits, and in an empirical investigation, the gain in accuracy of using ssGBLUP over BLUP increased as the deviations in phenotypic values of the animals increased. In addition, the superior accuracy of ssGBLUP tended to be more evident in animals whose maternal grandsire was genotyped than in other animals, although the effect was small.

Efficiency of genomic selection for breeding population design and phenotype prediction in tomato

Genomic selection (GS), which uses estimated genetic potential based on genome-wide genotype data for a breeding selection, is now widely accepted as an efficient method to improve genetically complex traits. We assessed the potential of GS for increasing soluble solids content and total fruit weight of tomato. A collection of big-fruited F1 varieties was used to construct the GS models, and the progeny from crosses was used to validate the models. The present study includes two experiments: a prediction of a parental combination that generates superior progeny and the prediction of progeny phenotypes. The GS models successfully predicted a better parent even if the phenotypic value did not vary substantially between candidates. The GS models also predicted phenotypes of progeny, although their efficiency varied depending on the parental cross combinations and the selected traits. Although further analyses are required to apply GS in an actual breeding situation, our results indicated that GS is a promising strategy for future tomato breeding design.

Uncovering a Nuisance Influence of a Phenological Trait of Plants Using a Nonlinear Structural Equation: Application to Days to Heading and Culm Length in Asian Cultivated Rice (Oryza Sativa L.).

Phenological traits of plants, such as flowering time, are linked to growth phase transition. Thus, phenological traits often influence other traits through the modification of the duration of growth period. This influence is a nuisance in plant breeding because it hampers genetic evaluation of the influenced traits. Genetic effects on the influenced traits have two components, one that directly affects the traits and one that indirectly affects the traits via the phenological trait. These cannot be distinguished by phenotypic evaluation and ordinary linear regression models. Consequently, if a phenological trait is modified by introgression or editing of the responsible genes, the phenotypes of the influenced traits can change unexpectedly. To uncover the influence of the phenological trait and evaluate the direct genetic effects on the influenced traits, we developed a nonlinear structural equation (NSE) incorporating a nonlinear influence of the phenological trait. We applied the NSE to real data for cultivated rice (Oryza sativa L.): days to heading (DH) as a phenological trait and culm length (CL) as the influenced trait. This showed that CL of the cultivars that showed extremely early heading was shortened by the strong influence of DH. In a simulation study, it was shown that the NSE was able to infer the nonlinear influence and direct genetic effects with reasonable accuracy. However, the NSE failed to infer the linear influence in this study. When no influence was simulated, an ordinary bi-trait linear model (OLM) tended to infer the genetic effects more accurately. In such cases, however, by comparing the NSE and OLM using an information criterion, we could assess whether the nonlinear assumption of the NSE was appropriate for the data analyzed. This study demonstrates the usefulness of the NSE in revealing the phenotypic influence of phenological traits.

Investigation of genetic diversity and inbreeding in a Japanese native horse breed for suggestions on its conservation

Because native breeds can serve as genetic resources for adapting to environment changes, their conservation is important for future agroecosystems. Using pedigree analysis, we investigated genetic diversity and inbreeding in Japanese Hokkaido native horses, which have adapted to a cold climate and roughage diet. Genetic diversity was measured as the number of founders and the effective number of founders, ancestors and genomes. All metrics imply a decrease in genetic diversity. A comparison of these metrics suggested that pedigree bottlenecks contributed more than did random gene losses to the reduction of genetic diversity. Estimates of marginal contributions of ancestors suggest that the bottlenecks arose mainly because related stallions had been used for breeding. A tendency for an increase in inbreeding coefficients was observed. F-statistics revealed that a small effective population size majorly contributed to this increase, although non-random mating in particular regions also contributed. Because the bottlenecks are thought to have reduced the effective population size, our results imply that mitigation of bottlenecks is important for conservation. To this end, breeding should involve genetically diverse stallions. In addition, to prevent non-random mating observed in particular regions, efforts should be made to plan mating with consideration of kinships.