Training Population Optimization for Genomic Selection in Miscanthus

Near-infrared reflectance spectroscopy phenomic prediction can perform similarly to genomic prediction of maize agronomic traits across environments

For nearly two decades, genomic prediction and selection have supported efforts to increase genetic gains in plant and animal improvement programs. However, novel phenomic strategies for predicting complex traits in maize have recently proven beneficial when integrated into across-environment sparse genomic prediction models. One phenomic data modality is whole grain near-infrared spectroscopy (NIRS), which records reflectance values of biological samples (e.g., maize kernels) based on chemical composition. Predictions of hybrid maize grain yield (GY) and 500-kernel weight (KW) across 2 years (2011-2012) and two management conditions (water-stressed and well-watered) were conducted using combinations of reflectance data obtained from high-throughput, F2 whole-kernel scans and genomic data obtained from genotyping-by-sequencing within four different cross-validation (CV) schemes (CV2, CV1, CV0, and CV00). When predicting the performance of untested genotypes in characterized (CV1) environments, genomic data were better than phenomic data for GY (0.689 ± 0.024-genomic vs. 0.612 ± 0.045-phenomic), but phenomic data were better than genomic data for KW (0.535 ± 0.034-genomic vs. 0.617 ± 0.145-phenomic). Multi-kernel models (combinations of phenomic and genomic relationship matrices) did not surpass single-kernel models for GY prediction in CV1 or CV00 (prediction of untested genotypes in uncharacterized environments); however, these models did outperform the single-kernel models for prediction of KW in these same CVs. Lasso regression applied to the NIRS data set selected a subset of 216 NIRS bands that achieved comparable prediction abilities to the full phenomic data set of 3112 bands predicting GY and KW under CV1 and CV00.

Genomic selection in plant breeding: Key factors shaping two decades of progress

Genomic selection, the application of genomic prediction (GP) models to select candidate individuals, has significantly advanced in the past two decades, effectively accelerating genetic gains in plant breeding. This article provides a holistic overview of key factors that have influenced GP in plant breeding during this period. We delved into the pivotal roles of training population size and genetic diversity, and their relationship with the breeding population, in determining GP accuracy. Special emphasis was placed on optimizing training population size. We explored its benefits and the associated diminishing returns beyond an optimum size. This was done while considering the balance between resource allocation and maximizing prediction accuracy through current optimization algorithms. The density and distribution of single-nucleotide polymorphisms, level of linkage disequilibrium, genetic complexity, trait heritability, statistical machine-learning methods, and non-additive effects are the other vital factors. Using wheat, maize, and potato as examples, we summarize the effect of these factors on the accuracy of GP for various traits. The search for high accuracy in GP-theoretically reaching one when using the Pearson's correlation as a metric-is an active research area as yet far from optimal for various traits. We hypothesize that with ultra-high sizes of genotypic and phenotypic datasets, effective training population optimization methods and support from other omics approaches (transcriptomics, metabolomics and proteomics) coupled with deep-learning algorithms could overcome the boundaries of current limitations to achieve the highest possible prediction accuracy, making genomic selection an effective tool in plant breeding.

Maximizing efficiency in sunflower breeding through historical data optimization

Genomic selection (GS) has become an increasingly popular tool in plant breeding programs, propelled by declining genotyping costs, an increase in computational power, and rediscovery of the best linear unbiased prediction methodology over the past two decades. This development has led to an accumulation of extensive historical datasets with genotypic and phenotypic information, triggering the question of how to best utilize these datasets. Here, we investigate whether all available data or a subset should be used to calibrate GS models for across-year predictions in a 7-year dataset of a commercial hybrid sunflower breeding program. We employed a multi-objective optimization approach to determine the ideal years to include in the training set (TRS). Next, for a given combination of TRS years, we further optimized the TRS size and its genetic composition. We developed the Min_GRM size optimization method which consistently found the optimal TRS size, reducing dimensionality by 20% with an approximately 1% loss in predictive ability. Additionally, the Tails_GEGVs algorithm displayed potential, outperforming the use of all data by using just 60% of it for grain yield, a high-complexity, low-heritability trait. Moreover, maximizing the genetic diversity of the TRS resulted in a consistent predictive ability across the entire range of genotypic values in the test set. Interestingly, the Tails_GEGVs algorithm, due to its ability to leverage heterogeneity, enhanced predictive performance for key hybrids with extreme genotypic values. Our study provides new insights into the optimal utilization of historical data in plant breeding programs, resulting in improved GS model predictive ability.

Accelerating genetic gains for quantitative resistance to verticillium wilt through predictive breeding in strawberry

Verticillium wilt (VW), a devastating vascular wilt disease of strawberry (Fragaria × $\times$ ananassa), has caused economic losses for nearly a century. This disease is caused by the soil-borne pathogen Verticillium dahliae, which occurs nearly worldwide and causes disease in numerous agriculturally important plants. The development of VW-resistant cultivars is critically important for the sustainability of strawberry production. We previously showed that a preponderance of the genetic resources (asexually propagated hybrid individuals) preserved in public germplasm collections were moderately to highly susceptible and that genetic gains for increased resistance to VW have been negligible over the last 60 years. To more fully understand the challenges associated with breeding for increased quantitative resistance to this pathogen, we developed and phenotyped a training population of hybrids (n = 564 $n = 564$ ) among elite parents with a wide range of resistance phenotypes. When these data were combined with training data from a population of elite and exotic hybrids (n = 386 $n = 386$ ), genomic prediction accuracies of 0.47-0.48 were achieved and were predicted to explain 70%-75% of the additive genetic variance for resistance. We concluded that breeding values for resistance to VW can be predicted with sufficient accuracy for effective genomic selection with routine updating of training populations.

Impact of genotype-calling methodologies on genome-wide association and genomic prediction in polyploids

Discovery and analysis of genetic variants underlying agriculturally important traits are key to molecular breeding of crops. Reduced representation approaches have provided cost-efficient genotyping using next-generation sequencing. However, accurate genotype calling from next-generation sequencing data is challenging, particularly in polyploid species due to their genome complexity. Recently developed Bayesian statistical methods implemented in available software packages, polyRAD, EBG, and updog, incorporate error rates and population parameters to accurately estimate allelic dosage across any ploidy. We used empirical and simulated data to evaluate the three Bayesian algorithms and demonstrated their impact on the power of genome-wide association study (GWAS) analysis and the accuracy of genomic prediction. We further incorporated uncertainty in allelic dosage estimation by testing continuous genotype calls and comparing their performance to discrete genotypes in GWAS and genomic prediction. We tested the genotype-calling methods using data from two autotetraploid species, Miscanthus sacchariflorus and Vaccinium corymbosum, and performed GWAS and genomic prediction. In the empirical study, the tested Bayesian genotype-calling algorithms differed in their downstream effects on GWAS and genomic prediction, with some showing advantages over others. Through subsequent simulation studies, we observed that at low read depth, polyRAD was advantageous in its effect on GWAS power and limit of false positives. Additionally, we found that continuous genotypes increased the accuracy of genomic prediction, by reducing genotyping error, particularly at low sequencing depth. Our results indicate that by using the Bayesian algorithm implemented in polyRAD and continuous genotypes, we can accurately and cost-efficiently implement GWAS and genomic prediction in polyploid crops.

A genome-wide association study and genomic prediction for Phakopsora pachyrhizi resistance in soybean

Soybean brown rust (SBR), caused by Phakopsora pachyrhizi, is a devastating fungal disease that threatens global soybean production. This study conducted a genome-wide association study (GWAS) with seven models on a panel of 3,082 soybean accessions to identify the markers associated with SBR resistance by 30,314 high quality single nucleotide polymorphism (SNPs). Then five genomic selection (GS) models, including Ridge regression best linear unbiased predictor (rrBLUP), Genomic best linear unbiased predictor (gBLUP), Bayesian least absolute shrinkage and selection operator (Bayesian LASSO), Random Forest (RF), and Support vector machines (SVM), were used to predict breeding values of SBR resistance using whole genome SNP sets and GWAS-based marker sets. Four SNPs, namely Gm18_57,223,391 (LOD = 2.69), Gm16_29,491,946 (LOD = 3.86), Gm06_45,035,185 (LOD = 4.74), and Gm18_51,994,200 (LOD = 3.60), were located near the reported P. pachyrhizi R genes, Rpp1, Rpp2, Rpp3, and Rpp4, respectively. Other significant SNPs, including Gm02_7,235,181 (LOD = 7.91), Gm02_7234594 (LOD = 7.61), Gm03_38,913,029 (LOD = 6.85), Gm04_46,003,059 (LOD = 6.03), Gm09_1,951,644 (LOD = 10.07), Gm10_39,142,024 (LOD = 7.12), Gm12_28,136,735 (LOD = 7.03), Gm13_16,350,701(LOD = 5.63), Gm14_6,185,611 (LOD = 5.51), and Gm19_44,734,953 (LOD = 6.02), were associated with abundant disease resistance genes, such as Glyma.02G084100, Glyma.03G175300, Glyma.04g189500, Glyma.09G023800, Glyma.12G160400, Glyma.13G064500, Glyma.14g073300, and Glyma.19G190200. The annotations of these genes included but not limited to: LRR class gene, cytochrome 450, cell wall structure, RCC1, NAC, ABC transporter, F-box domain, etc. The GWAS based markers showed more accuracies in genomic prediction than the whole genome SNPs, and Bayesian LASSO model was the ideal model in SBR resistance prediction with 44.5% ~ 60.4% accuracies. This study aids breeders in predicting selection accuracy of complex traits such as disease resistance and can shorten the soybean breeding cycle by the identified markers.

A comparison of methods for training population optimization in genomic selection

Maximizing CDmean and Avg_GRM_self were the best criteria for training set optimization. A training set size of 50-55% (targeted) or 65-85% (untargeted) is needed to obtain 95% of the accuracy.  With the advent of genomic selection (GS) as a widespread breeding tool, mechanisms to efficiently design an optimal training set for GS models became more relevant, since they allow maximizing the accuracy while minimizing the phenotyping costs. The literature described many training set optimization methods, but there is a lack of a comprehensive comparison among them. This work aimed to provide an extensive benchmark among optimization methods and optimal training set size by testing a wide range of them in seven datasets, six different species, different genetic architectures, population structure, heritabilities, and with several GS models to provide some guidelines about their application in breeding programs. Our results showed that targeted optimization (uses information from the test set) performed better than untargeted (does not use test set data), especially when heritability was low. The mean coefficient of determination was the best targeted method, although it was computationally intensive. Minimizing the average relationship within the training set was the best strategy for untargeted optimization. Regarding the optimal training set size, maximum accuracy was obtained when the training set was the entire candidate set. Nevertheless, a 50-55% of the candidate set was enough to reach 95-100% of the maximum accuracy in the targeted scenario, while we needed a 65-85% for untargeted optimization. Our results also suggested that a diverse training set makes GS robust against population structure, while including clustering information was less effective. The choice of the GS model did not have a significant influence on the prediction accuracies.

Building a Calibration Set for Genomic Prediction, Characteristics to Be Considered, and Optimization Approaches

The efficiency of genomic selection strongly depends on the prediction accuracy of the genetic merit of candidates. Numerous papers have shown that the composition of the calibration set is a key contributor to prediction accuracy. A poorly defined calibration set can result in low accuracies, whereas an optimized one can considerably increase accuracy compared to random sampling, for a same size. Alternatively, optimizing the calibration set can be a way of decreasing the costs of phenotyping by enabling similar levels of accuracy compared to random sampling but with fewer phenotypic units. We present here the different factors that have to be considered when designing a calibration set, and review the different criteria proposed in the literature. We classified these criteria into two groups: model-free criteria based on relatedness, and criteria derived from the linear mixed model. We introduce criteria targeting specific prediction objectives including the prediction of highly diverse panels, biparental families, or hybrids. We also review different ways of updating the calibration set, and different procedures for optimizing phenotyping experimental designs.

Genomics-Enabled Management of Genetic Resources in Radiata Pine

Traditional tree improvement is cumbersome and costly. Our main objective was to assess the extent to which genomic data can currently accelerate and improve decision making in this field. We used diameter at breast height (DBH) and wood density (WD) data for 4430 tree genotypes and single-nucleotide polymorphism (SNP) data for 2446 tree genotypes. Pedigree reconstruction was performed using a combination of maximum likelihood parentage assignment and matching based on identity-by-state (IBS) similarity. In addition, we used best linear unbiased prediction (BLUP) methods to predict phenotypes using SNP markers (GBLUP), recorded pedigree information (ABLUP), and single-step “blended” BLUP (HBLUP) combining SNP and pedigree information. We substantially improved the accuracy of pedigree records, resolving the inconsistent parental information of 506 tree genotypes. This led to substantially increased predictive ability (i.e., by up to 87%) in HBLUP analyses compared to a baseline from ABLUP. Genomic prediction was possible across populations and within previously untested families with moderately large training populations (N = 800–1200 tree genotypes) and using as few as 2000–5000 SNP markers. HBLUP was generally more effective than traditional ABLUP approaches, particularly after dealing appropriately with pedigree uncertainties. Our study provides evidence that genome-wide marker data can significantly enhance tree improvement. The operational implementation of genomic selection has started in radiata pine breeding in New Zealand, but further reductions in DNA extraction and genotyping costs may be required to realise the full potential of this approach.

Genetic characterization of carrot root shape and size using genome-wide association analysis and genomic-estimated breeding values

The principal phenotypic determinants of market class in carrot-the size and shape of the root-are under primarily additive, but also highly polygenic, genetic control. The size and shape of carrot roots are the primary determinants not only of yield, but also market class. These quantitative phenotypes have historically been challenging to objectively evaluate, and thus subjective visual assessment of market class remains the primary method by which selection for these traits is performed. However, advancements in digital image analysis have recently made possible the high-throughput quantification of size and shape attributes. It is therefore now feasible to utilize modern methods of genetic analysis to investigate the genetic control of root morphology. To this end, this study utilized both genome wide association analysis (GWAS) and genomic-estimated breeding values (GEBVs) and demonstrated that the components of market class are highly polygenic traits, likely under the influence of many small effect QTL. Relatively large proportions of additive genetic variance for many of the component phenotypes support high predictive ability of GEBVs; average prediction ability across underlying market class traits was 0.67. GWAS identified multiple QTL for four of the phenotypes which compose market class: length, aspect ratio, maximum width, and root fill, a previously uncharacterized trait which represents the size-independent portion of carrot root shape. By combining digital image analysis with GWAS and GEBVs, this study represents a novel advance in our understanding of the genetic control of market class in carrot. The immediate practical utility and viability of genomic selection for carrot market class is also described, and concrete guidelines for the design of training populations are provided.

Investigation of genetic relationships within three Miscanthus species using SNP markers identified with SLAF-seq

BACKGROUND

Miscanthus, which is a leading dedicated-energy grass in Europe and in parts of Asia, is expected to play a key role in the development of the future bioeconomy. However, due to its complex genetic background, it is difficult to investigate phylogenetic relationships in this genus. Here, we investigated 50 Miscanthus germplasms: 1 female parent (M. lutarioriparius), 30 candidate male parents (M. lutarioriparius, M. sinensis, and M. sacchariflorus), and 19 offspring. We used high-throughput Specific-Locus Amplified Fragment sequencing (SLAF-seq) to identify informative single nucleotide polymorphisms (SNPs) in all germplasms.

RESULTS

We identified 257,889 SLAF tags, of which 87,162 were polymorphic. Each tag was 264-364 bp long. The obtained 724,773 population SNPs were used to investigate genetic relationships within three species of Miscanthus. We constructed a phylogenetic tree of the 50 germplasms using the obtained SNPs and grouped them into two clades: one clade comprised of M. sinensis alone and the other one included the offspring, M. lutarioriparius, and M. sacchariflorus. Genetic cluster analysis had revealed that M. lutarioriparius germplasm C3 was the most likely male parent of the offspring.

CONCLUSIONS

As a high-throughput sequencing method, SLAF-seq can be used to identify informative SNPs in Miscanthus germplasms and to rapidly characterize genetic relationships within this genus. Our results will support the development of breeding programs with the focus on utilizing Miscanthus cultivars with elite biomass- or fiber-production potential for the developing bioeconomy.

Combining genetic resources and elite material populations to improve the accuracy of genomic prediction in apple

Genomic selection is an attractive strategy for apple breeding that could reduce the length of breeding cycles. A possible limitation to the practical implementation of this approach lies in the creation of a training set large and diverse enough to ensure accurate predictions. In this study, we investigated the potential of combining two available populations, i.e., genetic resources and elite material, in order to obtain a large training set with a high genetic diversity. We compared the predictive ability of genomic predictions within-population, across-population or when combining both populations, and tested a model accounting for population-specific marker effects in this last case. The obtained predictive abilities were moderate to high according to the studied trait and small increases in predictive ability could be obtained for some traits when the two populations were combined into a unique training set. We also investigated the potential of such a training set to predict hybrids resulting from crosses between the two populations, with a focus on the method to design the training set and the best proportion of each population to optimize predictions. The measured predictive abilities were very similar for all the proportions, except for the extreme cases where only one of the two populations was used in the training set, in which case predictive abilities could be lower than when using both populations. Using an optimization algorithm to choose the genotypes in the training set also led to higher predictive abilities than when the genotypes were chosen at random. Our results provide guidelines to initiate breeding programs that use genomic selection when the implementation of the training set is a limitation.

Genomic prediction and training set optimization in a structured Mediterranean oat population

The strong genetic structure observed in Mediterranean oats affects the predictive ability of genomic prediction as well as the performance of training set optimization methods. In this study, we investigated the efficiency of genomic prediction and training set optimization in a highly structured population of cultivars and landraces of cultivated oat (Avena sativa) from the Mediterranean basin, including white (subsp. sativa) and red (subsp. byzantina) oats, genotyped using genotype-by-sequencing markers and evaluated for agronomic traits in Southern Spain. For most traits, the predictive abilities were moderate to high with little differences between models, except for biomass for which Bayes-B showed a substantial gain compared to other models. The consistency between the structure of the training population and the population to be predicted was key to the predictive ability of genomic predictions. The predictive ability of inter-subspecies predictions was indeed much lower than that of intra-subspecies predictions for all traits. Regarding training set optimization, the linear mixed model optimization criteria (prediction error variance (PEVmean) and coefficient of determination (CDmean)) performed better than the heuristic approach "partitioning around medoids," even under high population structure. The superiority of CDmean and PEVmean could be explained by their ability to adapt the representation of each genetic group according to those represented in the population to be predicted. These results represent an important step towards the implementation of genomic prediction in oat breeding programs and address important issues faced by the genomic prediction community regarding population structure and training set optimization.

Experimental evaluation of genomic selection prediction for rust resistance in sugarcane

The total sugarcane (Saccharum L.) production has increased worldwide; however, the rate of growth is lower compared with other major crops, mainly due to a plateauing of genetic gain. Genomic selection (GS) has proven to substantially increase the rate of genetic gain in many crops. To investigate the utility of GS in future sugarcane breeding, a field trial was conducted using 432 sugarcane clones using an augmented design with two replications. Two major diseases in sugarcane, brown and orange rust (BR and OR), were screened artificially using whorl inoculation method in the field over two crop cycles. The genotypic data were generated through target enrichment sequencing technologies. After filtering, a set of 8,825 single nucleotide polymorphic markers were used to assess the prediction accuracy of multiple GS models. Using fivefold cross-validation, we observed GS prediction accuracies for BR and OR that ranged from 0.28 to 0.43 and 0.13 to 0.29, respectively, across two crop cycles and combined cycles. The prediction ability further improved by including a known major gene for resistance to BR as a fixed effect in the GS model. It also substantially reduced the minimum number of markers and training population size required for GS. The nonparametric GS models outperformed the parametric GS suggesting that nonadditive genetic effects could contribute genomic sources underlying BR and OR. This study demonstrated that GS could potentially predict the genomic estimated breeding value for selecting the desired germplasm for sugarcane breeding for disease resistance.

Increased Predictive Accuracy of Multi-Environment Genomic Prediction Model for Yield and Related Traits in Spring Wheat (Triticum aestivum L.)

Genomic selection (GS) has the potential to improve the selection gain for complex traits in crop breeding programs from resource-poor countries. The GS model performance in multi-environment (ME) trials was assessed for 141 advanced breeding lines under four field environments via cross-predictions. We compared prediction accuracy (PA) of two GS models with or without accounting for the environmental variation on four quantitative traits of significant importance, i.e., grain yield (GRYLD), thousand-grain weight, days to heading, and days to maturity, under North and Central Indian conditions. For each trait, we generated PA using the following two different ME cross-validation (CV) schemes representing actual breeding scenarios: (1) predicting untested lines in tested environments through the ME model (ME_CV1) and (2) predicting tested lines in untested environments through the ME model (ME_CV2). The ME predictions were compared with the baseline single-environment (SE) GS model (SE_CV1) representing a breeding scenario, where relationships and interactions are not leveraged across environments. Our results suggested that the ME models provide a clear advantage over SE models in terms of robust trait predictions. Both ME models provided 2-3 times higher prediction accuracies for all four traits across the four tested environments, highlighting the importance of accounting environmental variance in GS models. While the improvement in PA from SE to ME models was significant, the CV1 and CV2 schemes did not show any clear differences within ME, indicating the ME model was able to predict the untested environments and lines equally well. Overall, our results provide an important insight into the impact of environmental variation on GS in smaller breeding programs where these programs can potentially increase the rate of genetic gain by leveraging the ME wheat breeding trials.

Training Set Optimization for Sparse Phenotyping in Genomic Selection: A Conceptual Overview

Genomic selection (GS) is becoming an essential tool in breeding programs due to its role in increasing genetic gain per unit time. The design of the training set (TRS) in GS is one of the key steps in the implementation of GS in plant and animal breeding programs mainly because (i) TRS optimization is critical for the efficiency and effectiveness of GS, (ii) breeders test genotypes in multi-year and multi-location trials to select the best-performing ones. In this framework, TRS optimization can help to decrease the number of genotypes to be tested and, therefore, reduce phenotyping cost and time, and (iii) we can obtain better prediction accuracies from optimally selected TRS than an arbitrary TRS. Here, we concentrate the efforts on reviewing the lessons learned from TRS optimization studies and their impact on crop breeding and discuss important features for the success of TRS optimization under different scenarios. In this article, we review the lessons learned from training population optimization in plants and the major challenges associated with the optimization of GS including population size, the relationship between training and test set (TS), update of TRS, and the use of different packages and algorithms for TRS implementation in GS. Finally, we describe general guidelines to improving the rate of genetic improvement by maximizing the use of the TRS optimization in the GS framework.

TrainSel: An R Package for Selection of Training Populations

A major barrier to the wider use of supervised learning in emerging applications, such as genomic selection, is the lack of sufficient and representative labeled data to train prediction models. The amount and quality of labeled training data in many applications is usually limited and therefore careful selection of the training examples to be labeled can be useful for improving the accuracies in predictive learning tasks. In this paper, we present an R package, TrainSel, which provides flexible, efficient, and easy-to-use tools that can be used for the selection of training populations (STP). We illustrate its use, performance, and potentials in four different supervised learning applications within and outside of the plant breeding area.

Diversifying maize genomic selection models

Genomic selection (GS) is one of the most powerful tools available for maize breeding. Its use of genome-wide marker data to estimate breeding values translates to increased genetic gains with fewer breeding cycles. In this review, we cover the history of GS and highlight particular milestones during its adaptation to maize breeding. We discuss how GS can be applied to developing superior maize inbreds and hybrids. Additionally, we characterize refinements in GS models that could enable the encapsulation of non-additive genetic effects, genotype by environment interactions, and multiple levels of the biological hierarchy, all of which could ultimately result in more accurate predictions of breeding values. Finally, we suggest the stages in a maize breeding program where it would be beneficial to apply GS. Given the current sophistication of high-throughput phenotypic, genotypic, and other -omic level data currently available to the maize community, now is the time to explore the implications of their incorporation into GS models and thus ensure that genetic gains are being achieved as quickly and efficiently as possible.