Using visual scores for genomic prediction of complex traits in breeding programs.

KEY MESSAGE: An approach for handling visual scores with potential errors and subjectivity in scores was evaluated in simulated and blueberry recurrent selection breeding schemes to assist breeders in their decision-making. Most genomic prediction methods are based on assumptions of normality due to their simplicity and ease of implementation. However, in plant and animal breeding, continuous traits are often visually scored as categorical traits and analyzed as a Gaussian variable, thus violating the normality assumption, which could affect the prediction of breeding values and the estimation of genetic parameters. In this study, we examined the main challenges of visual scores for genomic prediction and genetic parameter estimation using mixed models, Bayesian, and machine learning methods. We evaluated these approaches using simulated and real breeding data sets. Our contribution in this study is a five-fold demonstration: (i) collecting data using an intermediate number of categories (1-3 and 1-5) is the best strategy, even considering errors associated with visual scores; (ii) Linear Mixed Models and Bayesian Linear Regression are robust to the normality violation, but marginal gains can be achieved when using Bayesian Ordinal Regression Models (BORM) and Random Forest Classification; (iii) genetic parameters are better estimated using BORM; (iv) our conclusions using simulated data are also applicable to real data in autotetraploid blueberry; and (v) a comparison of continuous and categorical phenotypes found that investing in the evaluation of 600-1000 categorical data points with low error, when it is not feasible to collect continuous phenotypes, is a strategy for improving predictive abilities. Our findings suggest the best approaches for effectively using visual scores traits to explore genetic information in breeding programs and highlight the importance of investing in the training of evaluator teams and in high-quality phenotyping.

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.

Genomic prediction applied to multiple traits and environments in second season maize hybrids.

Genomic selection has become a reality in plant breeding programs with the reduction in genotyping costs. Especially in maize breeding programs, it emerges as a promising tool for predicting hybrid performance. The dynamics of a commercial breeding program involve the evaluation of several traits simultaneously in a large set of target environments. Therefore, multi-trait multi-environment (MTME) genomic prediction models can leverage these datasets by exploring the correlation between traits and Genotype-by-Environment (G×E) interaction. Herein, we assess predictive abilities of univariate and multivariate genomic prediction models in a maize breeding program. To this end, we used data from 415 maize hybrids evaluated in 4 years of second season field trials for the traits grain yield, number of ears, and grain moisture. Genotypes of these hybrids were inferred in silico based on their parental inbred lines using single nucleotide polymorphisms (SNPs) markers obtained via genotyping-by-sequencing (GBS). Because genotypic information was available for only 257 hybrids, we used the genomic and pedigree relationship matrices to obtain the H matrix for all 415 hybrids. Our results demonstrated that in the single-environment context the use of multi-trait models was always superior in comparison to their univariate counterparts. Besides that, although MTME models were not particularly successful in predicting hybrid performance in untested years, they improved the ability to predict the performance of hybrids that had not been evaluated in any environment. However, the computational requirements of this kind of model could represent a limitation to its practical implementation and further investigation is necessary.

Priors for genotyping polyploids.

MOTIVATION: Empirical Bayes techniques to genotype polyploid organisms usually either (i) assume technical artifacts are known a priori or (ii) estimate technical artifacts simultaneously with the prior genotype distribution. Case (i) is unappealing as it places the onus on the researcher to estimate these artifacts, or to ensure that there are no systematic biases in the data. However, as we demonstrate with a few empirical examples, case (ii) makes choosing the class of prior genotype distributions extremely important. Choosing a class is either too flexible or too restrictive results in poor genotyping performance. RESULTS: We propose two classes of prior genotype distributions that are of intermediate levels of flexibility: the class of proportional normal distributions and the class of unimodal distributions. We provide a complete characterization of and optimization details for the class of unimodal distributions. We demonstrate, using both simulated and real data that using these classes results in superior genotyping performance. AVAILABILITY AND IMPLEMENTATION: Genotyping methods that use these priors are implemented in the updog R package available on the Comprehensive R Archive Network: https://cran.r-project.org/package=updog. All code needed to reproduce the results of this article is available on GitHub: https://github.com/dcgerard/reproduce_prior_sims. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Accurate genomic prediction of Coffea canephora in multiple environments using whole-genome statistical models.

Genomic selection has been proposed as the standard method to predict breeding values in animal and plant breeding. Although some crops have benefited from this methodology, studies in Coffea are still emerging. To date, there have been no studies describing how well genomic prediction models work across populations and environments for different complex traits in coffee. Considering that predictive models are based on biological and statistical assumptions, it is expected that their performance vary depending on how well these assumptions align with the true genetic architecture of the phenotype. To investigate this, we used data from two recurrent selection populations of Coffea canephora, evaluated in two locations, and single nucleotide polymorphisms identified by Genotyping-by-Sequencing. In particular, we evaluated the performance of 13 statistical approaches to predict three important traits in the coffee-production of coffee beans, leaf rust incidence and yield of green beans. Analyses were performed for predictions within-environment, across locations and across populations to assess the reliability of genomic selection. Overall, differences in the prediction accuracy of the competing models were small, although the Bayesian methods showed a modest improvement over other methods, at the cost of more computation time. As expected, predictive accuracy for within-environment analysis, on average, were higher than predictions across locations and across populations. Our results support the potential of genomic selection to reshape traditional plant breeding schemes. In practice, we expect to increase the genetic gain per unit of time by reducing the length cycle of recurrent selection in coffee.

Genotyping Polyploids from Messy Sequencing Data.

Detecting and quantifying the differences in individual genomes (i.e., genotyping), plays a fundamental role in most modern bioinformatics pipelines. Many scientists now use reduced representation next-generation sequencing (NGS) approaches for genotyping. Genotyping diploid individuals using NGS is a well-studied field, and similar methods for polyploid individuals are just emerging. However, there are many aspects of NGS data, particularly in polyploids, that remain unexplored by most methods. Our contributions in this paper are fourfold: (i) We draw attention to, and then model, common aspects of NGS data: sequencing error, allelic bias, overdispersion, and outlying observations. (ii) Many datasets feature related individuals, and so we use the structure of Mendelian segregation to build an empirical Bayes approach for genotyping polyploid individuals. (iii) We develop novel models to account for preferential pairing of chromosomes, and harness these for genotyping. (iv) We derive oracle genotyping error rates that may be used for read depth suggestions. We assess the accuracy of our method in simulations, and apply it to a dataset of hexaploid sweet potato (Ipomoea batatas). An R package implementing our method is available at https://cran.r-project.org/package=updog.