bWGR: Bayesian Whole-Genome Regression

Megavariate methods capture complex genotype-by-environment interactions

Genomic prediction models that capture genotype-by-environment (GxE) interaction are useful for predicting site-specific performance by leveraging information among related individuals and correlated environments, but implementing such models is computationally challenging. This study describes the algorithm of these scalable approaches, including 2 models with latent representations of GxE interactions, namely MegaLMM and MegaSEM, and an efficient multivariate mixed-model solver, namely Pseudo-expectation Gauss-Seidel (PEGS), fitting different covariance structures [unstructured, extended factor analytic (XFA), Heteroskedastic compound symmetry (HCS)]. Accuracy and runtime are benchmarked on simulated scenarios with varying numbers of genotypes and environments. MegaLMM and PEGS-based XFA and HCS models provided the highest accuracy under sparse testing with 100 testing environments. PEGS-based unstructured model was orders of magnitude faster than restricted maximum likelihood (REML) based multivariate genomic best linear unbiased predictions (GBLUP) while providing the same accuracy. MegaSEM provided the lowest runtime, fitting a model with 200 traits and 20,000 individuals in ∼5 min, and a model with 2,000 traits and 2,000 individuals in less than 3 min. With the genomes-to-fields data, the most accurate predictions were attained with the univariate model fitted across environments and by averaging environment-level genomic estimated breeding values (GEBVs) from models with HCS and XFA covariance structures.

Global genotype by environment prediction competition reveals that diverse modeling strategies can deliver satisfactory maize yield estimates

Predicting phenotypes from a combination of genetic and environmental factors is a grand challenge of modern biology. Slight improvements in this area have the potential to save lives, improve food and fuel security, permit better care of the planet, and create other positive outcomes. In 2022 and 2023, the first open-to-the-public Genomes to Fields initiative Genotype by Environment prediction competition was held using a large dataset including genomic variation, phenotype and weather measurements, and field management notes gathered by the project over 9 years. The competition attracted registrants from around the world with representation from academic, government, industry, and nonprofit institutions as well as unaffiliated. These participants came from diverse disciplines, including plant science, animal science, breeding, statistics, computational biology, and others. Some participants had no formal genetics or plant-related training, and some were just beginning their graduate education. The teams applied varied methods and strategies, providing a wealth of modeling knowledge based on a common dataset. The winner's strategy involved 2 models combining machine learning and traditional breeding tools: 1 model emphasized environment using features extracted by random forest, ridge regression, and least squares, and 1 focused on genetics. Other high-performing teams' methods included quantitative genetics, machine learning/deep learning, mechanistic models, and model ensembles. The dataset factors used, such as genetics, weather, and management data, were also diverse, demonstrating that no single model or strategy is far superior to all others within the context of this competition.

Quantitative Genetics of Microbiome Mediated Traits

Multicellular organisms host a rich assemblage of associated microorganisms, collectively known as their "microbiomes". Microbiomes have the capacity to influence their hosts' fitnesses, but the conditions under which such influences contribute to evolution are not clear. This is due in part to a lack of a comprehensive theoretical framework for describing the combined effects of host and associated microbes on phenotypic variation. Here we begin to address this gap by extending the foundations of quantitative genetic theory to include host-associated microbes, as well as alleles of hosts, as factors that explain quantitative host trait variation. We introduce a way to partition host-associated microbiomes into componenents relevant for predicting a microbiome-mediated response to selection. We then apply our general framework to a simulation model of microbiome inheritance to illustrate principles for predicting host trait dynamics, and to generalize classical narrow and broad sense heritabilities to account for microbial effects. We demonstrate that microbiome-mediated responses to host selection can arise from various transmission modes, not solely vertical, with the contribution of non-vertical modes depending on host life history. Our work lays a foundation for integrating microbiome-mediated host variation and adaptation into our understanding of natural variation.

Global Genotype by Environment Prediction Competition Reveals That Diverse Modeling Strategies Can Deliver Satisfactory Maize Yield Estimates

Predicting phenotypes from a combination of genetic and environmental factors is a grand challenge of modern biology. Slight improvements in this area have the potential to save lives, improve food and fuel security, permit better care of the planet, and create other positive outcomes. In 2022 and 2023 the first open-to-the-public Genomes to Fields (G2F) initiative Genotype by Environment (GxE) prediction competition was held using a large dataset including genomic variation, phenotype and weather measurements and field management notes, gathered by the project over nine years. The competition attracted registrants from around the world with representation from academic, government, industry, and non-profit institutions as well as unaffiliated. These participants came from diverse disciplines include plant science, animal science, breeding, statistics, computational biology and others. Some participants had no formal genetics or plant-related training, and some were just beginning their graduate education. The teams applied varied methods and strategies, providing a wealth of modeling knowledge based on a common dataset. The winner's strategy involved two models combining machine learning and traditional breeding tools: one model emphasized environment using features extracted by Random Forest, Ridge Regression and Least-squares, and one focused on genetics. Other high-performing teams' methods included quantitative genetics, classical machine learning/deep learning, mechanistic models, and model ensembles. The dataset factors used, such as genetics; weather; and management data, were also diverse, demonstrating that no single model or strategy is far superior to all others within the context of this competition.

Integrating Bioinformatics and Machine Learning for Genomic Prediction in Chickens

Genomic prediction plays an increasingly important role in modern animal breeding, with predictive accuracy being a crucial aspect. The classical linear mixed model is gradually unable to accommodate the growing number of target traits and the increasingly intricate genetic regulatory patterns. Hence, novel approaches are necessary for future genomic prediction. In this study, we used an illumina 50K SNP chip to genotype 4190 egg-type female Rhode Island Red chickens. Machine learning (ML) and classical bioinformatics methods were integrated to fit genotypes with 10 economic traits in chickens. We evaluated the effectiveness of ML methods using Pearson correlation coefficients and the RMSE between predicted and actual phenotypic values and compared them with rrBLUP and BayesA. Our results indicated that ML algorithms exhibit significantly superior performance to rrBLUP and BayesA in predicting body weight and eggshell strength traits. Conversely, rrBLUP and BayesA demonstrated 2-58% higher predictive accuracy in predicting egg numbers. Additionally, the incorporation of suggestively significant SNPs obtained through the GWAS into the ML models resulted in an increase in the predictive accuracy of 0.1-27% across nearly all traits. These findings suggest the potential of combining classical bioinformatics methods with ML techniques to improve genomic prediction in the future.

Genetic loci associated with tissue-specific resistance to powdery mildew in octoploid strawberry (Fragaria × ananassa)

Powdery mildew is one of the most problematic diseases in strawberry production. To date, few commercial strawberry cultivars are deemed to have complete resistance and as such, an extensive spray programme must be implemented to control the pathogen. Here, a large-scale field experiment was used to determine the powdery mildew resistance status of leaf and fruit tissues across a diverse panel of strawberry genotypes. This phenotypic data was used to identify Quantitative Trait Nucleotides (QTN) associated with tissue-specific powdery mildew resistance. In total, six stable QTN were found to be associated with foliar resistance, with one QTN on chromosome 7D associated with a 61% increase in resistance. In contrast to the foliage results, there were no QTN associated with fruit disease resistance and there was a high level of resistance observed on strawberry fruit, with no genetic correlation observed between fruit and foliar symptoms, indicating a tissue-specific response. Beyond the identification of genetic loci, we also demonstrate that genomic selection can lead to rapid gains in foliar resistance across genotypes, with the potential to capture >50% of the genetic foliage resistance present in the population. To date, breeding of robust powdery mildew resistance in strawberry has been impeded by the quantitative nature of natural resistance and a lack of knowledge relating to the genetic control of the trait. These results address this shortfall, through providing the community with a wealth of information that could be utilized for genomic informed breeding, implementation of which could deliver a natural resistance strategy for combatting powdery mildew.

Genomic selection of soybean (Glycine max) for genetic improvement of yield and seed composition in a breeding context

Genomic selection has been utilized for genetic improvement in both plant and animal breeding and is a favorable technique for quantitative trait development. Within this study, genomic selection was evaluated within a breeding program, using novel validation methods in addition to plant materials and data from a commercial soybean (Glycine max) breeding program. A total of 1501 inbred lines were used to test multiple genomic selection models for multiple traits. Validation included cross-validation, inter-environment, and empirical validation. The results indicated that the extended genomic best linear unbiased prediction (EGBLUP) model was the most effective model tested for yield, protein, and oil in cross-validation with accuracies of 0.50, 0.68, and 0.64, respectively. Increasing marker number from 1000 to 3000 to 6000 single nucleotide polymorphism markers leads to statistically significant increases in accuracy. Cross-environment predictions were statistically lower than cross-validation with accuracies of 0.24, 0.54, and 0.42 for yield, protein, and oil, respectively, using the extended genomic BLUP model. Empirical validation, predicting the yield of 510 soybean lines, had a prediction accuracy of 0.34, with the inclusion of a maturity covariate leading to a notable increase in accuracy. Genomic selection identified high-performance lines in inter-environment predictions: 34% of lines within the upper quartile of yield, and 51% and 48% of the highest quartile protein and oil lines, respectively. Statistically similar results occurred comparing rankings in empirical validation and selection for advancements in yield trials. These results indicate that genomic selection is a useful tool for selection decisions.

Identification of potential genetic Loci and polygenic risk model for Budd-Chiari syndrome in Chinese population

Budd-Chiari syndrome (BCS) is characterized by hepatic venous outflow obstruction, posing life-threatening risks in severe cases. Reported risk factors include inherited and acquired hypercoagulable states or other predisposing factors. However, many patients have no identifiable etiology, and causes of BCS differ between the West and East. This study recruited 500 BCS patients and 696 normal individuals for whole-exome sequencing and developed a polygenic risk scoring (PRS) model using PLINK, LASSOSUM, BLUP, and BayesA methods. Risk factors for venous thromboembolism and vascular malformations were also assessed for BCS risk prediction. Ultimately, we discovered potential BCS risk mutations, such as rs1042331, and the optimal BayesA-generated PRS model presented an AUC >0.9 in the external replication cohort. This model provides particular insights into genetic risk differences between China and the West and suggests shared genetic risks among BCS, venous thromboembolism, and vascular malformations, offering different perspectives on BCS pathogenesis.

Genomic prediction of optimal cross combinations to accelerate genetic improvement of soybean (Glycine max)

Improving yield is a primary soybean breeding goal, as yield is the main determinant of soybean's profitability. Within the breeding process, selection of cross combinations is one of most important elements. Cross prediction will assist soybean breeders in identifying the best cross combinations among parental genotypes prior to crossing, increasing genetic gain and breeding efficiency. In this study optimal cross selection methods were created and applied in soybean and validated using historical data from the University of Georgia soybean breeding program, under multiple training set compositions and marker densities utilizing multiple genomic selection models for marker evaluation. Plant materials consisted of 702 advanced breeding lines evaluated in multiple environments and genotyped using SoySNP6k BeadChips. An additional marker set, the SoySNP3k marker set, was tested in this study as well. Optimal cross selection methods were used to predict the yield of 42 previously made crosses and compared to the performance of the cross's offspring in replicated field trials. The best prediction accuracy was obtained when using Extended Genomic BLUP with the SoySNP6k marker set, consisting of 3,762 polymorphic markers, with an accuracy of 0.56 with a training set maximally related to the crosses predicted and 0.4 in a training set with minimized relatedness to predicted crosses. Prediction accuracy was most significantly impacted by training set relatedness to the predicted crosses, marker density, and the genomic model used to predict marker effects. The usefulness criterion selected had an impact on prediction accuracy within training sets with low relatedness to the crosses predicted. Optimal cross prediction provides a useful method that assists plant breeders in selecting crosses in soybean breeding.

Physiological breeding for yield improvement in soybean: solar radiation interception-conversion, and harvest index

KEY MESSAGE

Efficiency of light interception, Radiation use efficiency and harvest index can be used as targets to improve grain yield potential in soybean. Grain yield (GY) production can be expressed as the result of three main efficiencies: light interception (Ei), radiation use (RUE), and harvest index (HI). Although dissecting GY through these three efficiencies is not entirely new, there is a lack of knowledge about the phenotypic variation, the genetic architecture, and the relative contribution of these three efficiencies on GY in soybean. This knowledge gap coupled with laborious phenotyping prevents the active consideration of these efficiencies into breeding programs. This study aims to reveal the phenotypic variation, heritability, genetic relationships, genetic architecture, and genomic prediction for Ei, RUE, and HI in soybean. We evaluated a maturity control panel of 383 Recombinant Inbred Lines (RILs) selected from the soybean nested association mapping (SoyNAM) population. Dry matter ground measured along with canopy coverage (CC) from UAS imagery were collected in three environments. Light interception was modeled through a logistic curve using CC as a proxy. The total above-ground biomass collected during the growing season and its respective cumulative light intercepted were used to derive RUE through linear models fitting. Additive-genetic correlations, genome-wide association (GWA) and whole-genome regressions (WGR) were performed to evaluate the relationship between traits, their association with genomic regions, and the feasibility of predicting these efficiencies with genomic information. Correlation analyses considered three groups: the entire data set, and the high- and low-yielding RILs to determine association as a function of the GY. Our results revealed moderate to high phenotypic variation for Ei, RUE, and HI with ranges of 8.5%, 1.1 g MJ-1, and 0.2, respectively. Additive-genetic correlation revealed a strong relationship of GY with HI and moderate with RUE and Ei when whole data set was considered, but negligible contribution of HI on GY when just the top 100 was analyzed. The GWA analyses showed that Ei is associated with three SNPs; two of them located on chromosome 7 and one on chromosome 11 with no previous quantitative trait loci (QTLs) reported for these regions. RUE is associated with four SNPs on chromosomes 1, 7, 11, and 18. Some of these QTLs are novel, while others are previously documented for plant architecture and chlorophyll content. Two SNPs positioned on chromosome 13 and 15 with previous QTLs reported for plant height and seed set, weight and abortion were associated with HI. WGR showed high predictive ability for Ei, RUE, and HI with maximum correlation ranging between 0.75 and 0.80. Future improvements in GY can be expected through strategies prioritizing Ei for short-term results when using high yielding germplasm and RUE for medium- and long-term outcomes. This work is a pioneer attempt to integrate traditional physiological traits into the breeding process in the context of physiological breeding.

FMixFN: A Fast Big Data-Oriented Genomic Selection Model Based on an Iterative Conditional Expectation algorithm

Genomic selection is an approach to select elite breeding stock based on the use of dense genetic markers and that has led to the development of various models to derive a predictive equation. However, the current genomic selection software faces several issues such as low prediction accuracy, low computational efficiency, or an inability to handle large-scale sample data. We report the development of a genomic prediction model named FMixFN with four zero-mean normal distributions as the prior distributions to optimize the predictive ability and computing efficiency. The variance of the prior distributions in our model is precisely determined based on an F2 population, and genomic estimated breeding values (GEBV) can be obtained accurately and quickly in combination with an iterative conditional expectation algorithm. We demonstrated that FMixFN improves computational efficiency and predictive ability compared to other methods, such as GBLUP, SSgblup, MIX, BayesR, BayesA, and BayesB. Most importantly, FMixFN may handle large-scale sample data, and thus should be able to meet the needs of large breeding companies or combined breeding schedules. Our study developed a Bayes genomic selection model called FMixFN, which combines stable predictive ability and high computational efficiency, and is a big data-oriented genomic selection model that has potential in the future. The FMixFN method can be freely accessed at https://zenodo.org/record/5560913 (DOI: 10.5281/zenodo.5560913).

Impact of Genomic Prediction Model, Selection Intensity, and Breeding Strategy on the Long-Term Genetic Gain and Genetic Erosion in Soybean Breeding

Genomic-assisted breeding has become an important tool in soybean breeding. However, the impact of different genomic selection (GS) approaches on short- and long-term gains is not well understood. Such gains are conditional on the breeding design and may vary with a combination of the prediction model, family size, selection strategies, and selection intensity. To address these open questions, we evaluated various scenarios through a simulated closed soybean breeding program over 200 breeding cycles. Genomic prediction was performed using genomic best linear unbiased prediction (GBLUP), Bayesian methods, and random forest, benchmarked against selection on phenotypic values, true breeding values (TBV), and random selection. Breeding strategies included selections within family (WF), across family (AF), and within pre-selected families (WPSF), with selection intensities of 2.5, 5.0, 7.5, and 10.0%. Selections were performed at the F4 generation, where individuals were phenotyped and genotyped with a 6K single nucleotide polymorphism (SNP) array. Initial genetic parameters for the simulation were estimated from the SoyNAM population. WF selections provided the most significant long-term genetic gains. GBLUP and Bayesian methods outperformed random forest and provided most of the genetic gains within the first 100 generations, being outperformed by phenotypic selection after generation 100. All methods provided similar performances under WPSF selections. A faster decay in genetic variance was observed when individuals were selected AF and WPSF, as 80% of the genetic variance was depleted within 28-58 cycles, whereas WF selections preserved the variance up to cycle 184. Surprisingly, the selection intensity had less impact on long-term gains than did the breeding strategies. The study supports that genetic gains can be optimized in the long term with specific combinations of prediction models, family size, selection strategies, and selection intensity. A combination of strategies may be necessary for balancing the short-, medium-, and long-term genetic gains in breeding programs while preserving the genetic variance.

Genomic prediction using training population design in interspecific soybean populations

Agronomically important traits generally have complex genetic architecture, where many genes have a small and largely additive effect. Genomic prediction has been demonstrated to increase genetic gain and efficiency in plant breeding programs beyond marker-assisted selection and phenotypic selection. The objective of this study was to evaluate the impact of allelic origin, marker density, training population size, and cross-validation schemes on the accuracy of genomic prediction models in an interspecific soybean nested association mapping (NAM) panel. Three cross-validation schemes were used: (a) Within-Family (WF): training population and predictions are made exclusively within each family; (b) Across All families (AF): all the individuals from the three families were randomly assigned to either the training or validation set; (c) Leave one Family out (LFO): each family is predicted using a training set that contains the other two families. Predictive abilities increased with training population size up to 350 individuals, but no significant gains were noted beyond 250 individuals in the training population. The number of markers had a limited impact on the observed predictive ability across traits; increasing markers used in the model above 1000 revealed no significant increases in prediction accuracy. Predictive abilities for AF were not significantly different from the WF method, and predictive abilities across populations for the WF method had a range of 0.58 to 0.70 for maturity, protein, meal, and oil. Our results also showed encouraging prediction accuracies for grain yield (0.58-0.69) using the WF method. Partitioning genomic prediction between G. max and G. soja alleles revealed useful information to select material with a larger allele contribution from both parents and could accelerate allele introgression from exotic germplasm into the elite soybean gene pool.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11032-021-01203-6.

Quantitative Genomic Dissection of Soybean Yield Components

Soybean is a crop of major economic importance with low rates of genetic gains for grain yield compared to other field crops. A deeper understanding of the genetic architecture of yield components may enable better ways to tackle the breeding challenges. Key yield components include the total number of pods, nodes and the ratio pods per node. We evaluated the SoyNAM population, containing approximately 5600 lines from 40 biparental families that share a common parent, in 6 environments distributed across 3 years. The study indicates that the yield components under evaluation have low heritability, a reasonable amount of epistatic control, and partially oligogenic architecture: 18 quantitative trait loci were identified across the three yield components using multi-approach signal detection. Genetic correlation between yield and yield components was highly variable from family-to-family, ranging from -0.2 to 0.5. The genotype-by-environment correlation of yield components ranged from -0.1 to 0.4 within families. The number of pods can be utilized for indirect selection of yield. The selection of soybean for enhanced yield components can be successfully performed via genomic prediction, but the challenging data collections necessary to recalibrate models over time makes the introgression of QTL a potentially more feasible breeding strategy. The genomic prediction of yield components was relatively accurate across families, but less accurate predictions were obtained from within family predictions and predicting families not observed included in the calibration set.