Accuracies of univariate and multivariate genomic prediction models in African cassava

Enhancing the potential of phenomic and genomic prediction in winter wheat breeding using high-throughput phenotyping and deep learning

Integrating high-throughput phenotyping (HTP) based traits into phenomic and genomic selection (GS) can accelerate the breeding of high-yielding and climate-resilient wheat cultivars. In this study, we explored the applicability of Unmanned Aerial Vehicles (UAV)-assisted HTP combined with deep learning (DL) for the phenomic or multi-trait (MT) genomic prediction of grain yield (GY), test weight (TW), and grain protein content (GPC) in winter wheat. Significant correlations were observed between agronomic traits and HTP-based traits across different growth stages of winter wheat. Using a deep neural network (DNN) model, HTP-based phenomic predictions showed robust prediction accuracies for GY, TW, and GPC for a single location with R2 of 0.71, 0.62, and 0.49, respectively. Further prediction accuracies increased (R2 of 0.76, 0.64, and 0.75) for GY, TW, and GPC, respectively when advanced breeding lines from multi-locations were used in the DNN model. Prediction accuracies for GY varied across growth stages, with the highest accuracy at the Feekes 11 (Milky ripe) stage. Furthermore, forward prediction of GY in preliminary breeding lines using DNN trained on multi-location data from advanced breeding lines improved the prediction accuracy by 32% compared to single-location data. Next, we evaluated the potential of incorporating HTP-based traits in multi-trait genomic selection (MT-GS) models in the prediction of GY, TW, and GPC. MT-GS, models including UAV data-based anthocyanin reflectance index (ARI), green chlorophyll index (GCI), and ratio vegetation index 2 (RVI_2) as covariates demonstrated higher predictive ability (0.40, 0.40, and 0.37, respectively) as compared to single-trait model (0.23) for GY. Overall, this study demonstrates the potential of integrating HTP traits into DL-based phenomic or MT-GS models for enhancing breeding efficiency.

Multi-trait selection in multi-environments for performance and stability in cassava genotypes

Genotype-environment interaction (GEI) presents challenges when aiming to select optimal cassava genotypes, often due to biased genetic estimates. Various strategies have been proposed to address the need for simultaneous improvements in multiple traits, while accounting for performance and yield stability. Among these methods are mean performance and stability (MPS) and the multi-trait mean performance and stability index (MTMPS), both utilizing linear mixed models. This study's objective was to assess genetic variation and GEI effects on fresh root yield (FRY), along with three primary and three secondary traits. A comprehensive evaluation of 22 genotypes was conducted using a randomized complete block design with three replicates across 47 distinct environments (year x location) in Brazil. The broad-sense heritability (H 2 ) averaged 0.37 for primary traits and 0.44 for secondary traits, with plot-based heritability (h m ɡ 2 ) consistently exceeding 0.90 for all traits. The high extent of GEI variance (σ ɡ x e 2 ) demonstrates the GEI effect on the expression of these traits. The dominant analytic factor ( F A 3 ) accounted for over 85% of the total variance, and the communality (ɧ) surpassed 87% for all traits. These values collectively suggest a substantial capacity for genetic variance explanation. In Cluster 1, composed of remarkably productive and stable genotypes for primary traits, genotypes BRS Novo Horizonte and BR11-34-69 emerged as prime candidates for FRY enhancement, while BRS Novo Horizonte and BR12-107-002 were indicated for optimizing dry matter content. Moreover, MTMPS, employing a selection intensity of 30%, identified seven genotypes distinguished by heightened stability. This selection encompassed innovative genotypes chosen based on regression variance index (S d i 2 , R 2 , and RMSE) considerations for multiple traits. In essence, incorporating methodologies that account for stability and productive performance can significantly bolster the credibility of recommendations for novel cassava cultivars.

Improved multi-trait prediction of wheat end-product quality traits by integrating NIR-predicted phenotypes

Historically, end-product quality testing has been costly and required large flour samples; therefore, it was generally implemented in the late phases of variety development, imposing a huge cost on the breeding effort and effectiveness. High genetic correlations of end-product quality traits with higher throughput and nondestructive testing technologies, such as near-infrared (NIR), could enable early-stage testing and effective selection of these highly valuable traits in a multi-trait genomic prediction model. We studied the impact on prediction accuracy in genomic best linear unbiased prediction (GBLUP) of adding NIR-predicted secondary traits for six end-product quality traits (crumb yellowness, water absorption, texture hardness, flour yield, grain protein, flour swelling volume). Bread wheat lines (1,400-1,900) were measured across 8 years (2012-2019) for six end-product quality traits with standard laboratory assays and with NIR, which were combined to generate predicted data for approximately 27,000 lines. All lines were genotyped with the Infinium™ Wheat Barley 40K BeadChip and imputed using exome sequence data. End-product and NIR phenotypes were genetically correlated (0.5-0.83, except for flour swelling volume 0.19). Prediction accuracies of end-product traits ranged between 0.28 and 0.64 and increased by 30% through the inclusion of NIR-predicted data compared to single-trait analysis. There was a high correlation between the multi-trait prediction accuracy and genetic correlations between end-product and NIR-predicted data (0.69-0.77). Our forward prediction validation revealed a gradual increase in prediction accuracy when adding more years to the multi-trait model. Overall, we achieved genomic prediction accuracy at a level that enables selection for end-product quality traits early in the breeding cycle.

Effects of common full-sib families on accuracy of genomic prediction for tagging weight in striped catfish Pangasianodon hypophthalmus

Common full-sib families (c ² ) make up a substantial proportion of total phenotypic variation in traits of commercial importance in aquaculture species and omission or inclusion of the c ² resulted in possible changes in genetic parameter estimates and re-ranking of estimated breeding values. However, the impacts of common full-sib families on accuracy of genomic prediction for commercial traits of economic importance are not well known in many species, including aquatic animals. This research explored the impacts of common full-sib families on accuracy of genomic prediction for tagging weight in a population of striped catfish comprising 11,918 fish traced back to the base population (four generations), in which 560 individuals had genotype records of 14,154 SNPs. Our single step genomic best linear unbiased prediction (ssGLBUP) showed that the accuracy of genomic prediction for tagging weight was reduced by 96.5%-130.3% when the common full-sib families were included in statistical models. The reduction in the prediction accuracy was to a smaller extent in multivariate analysis than in univariate models. Imputation of missing genotypes somewhat reduced the upward biases in the prediction accuracy for tagging weight. It is therefore suggested that genomic evaluation models for traits recorded during the early phase of growth development should account for the common full-sib families to minimise possible biases in the accuracy of genomic prediction and hence, selection response.

Validation of KASP-SNP markers in cassava germplasm for marker-assisted selection of increased carotenoid content and dry matter content

Provitamin A biofortification and increased dry matter content are important breeding targets in cassava improvement programs worldwide. Biofortified varieties contribute to the alleviation of provitamin A deficiency, a leading cause of preventable blindness common among pre-school children and pregnant women in developing countries particularly Africa. Dry matter content is a major component of dry yield and thus underlies overall variety performance and acceptability by growers, processors, and consumers. Single nucleotide polymorphism (SNP) markers linked to these traits have recently been discovered through several genome-wide association studies but have not been deployed for routine marker-assisted selection (MAS). This is due to the lack of useful information on markers' performances in diverse genetic backgrounds. To overcome this bottleneck, technical and biological validation of the loci associated with increased carotenoid content and dry matter content were carried out using populations independent of the marker discovery population. In the present study, seven previously identified markers for these traits were converted to a robust set of uniplex allele-specific polymerase chain reaction (PCR) assays and validated in two independent pre-breeding and breeding populations. These assays were efficient in discriminating marker genotypic classes and had an average call rate greater than 98%. A high correlation was observed between the predicted and observed carotenoid content as inferred by root yellowness intensity in the breeding (r = 0.92) and pre-breeding (r = 0.95) populations. On the other hand, dry matter content-markers had moderately low predictive accuracy in both populations (r< 0.40) due to the more quantitative nature of the trait. This work confirmed the markers' effectiveness in multiple backgrounds, therefore, further strengthening their value in cassava biofortification to ensure nutritional security as well as dry matter content productivity. Our study provides a framework to guide future marker validation, thus leading to the more routine use of markers in MAS in cassava improvement programs.

Multi-trait genome prediction of new environments with partial least squares

The genomic selection (GS) methodology proposed over 20 years ago by Meuwissen et al. (Genetics, 2001) has revolutionized plant breeding. A predictive methodology that trains statistical machine learning algorithms with phenotypic and genotypic data of a reference population and makes predictions for genotyped candidate lines, GS saves significant resources in the selection of candidate individuals. However, its practical implementation is still challenging when the plant breeder is interested in the prediction of future seasons or new locations and/or environments, which is called the "leave one environment out" issue. Furthermore, because the distributions of the training and testing set do not match, most statistical machine learning methods struggle to produce moderate or reasonable prediction accuracies. For this reason, the main objective of this study was to explore the use of the multi-trait partial least square (MT-PLS) regression methodology for this specific task, benchmarking its performance with the Bayesian Multi-trait Genomic Best Linear Unbiased Predictor (MT-GBLUP) method. The benchmarking process was performed with five actual data sets. We found that in all data sets the MT-PLS method outperformed the popular MT-GBLUP method by 349.8% (under predictor E + G), 484.4% (under predictor E + G + GE; where E denotes environments, G genotypes and GE the genotype by environment interaction) and 15.9% (under predictor G + GE) across traits. Our results provide empirical evidence of the power of the MT-PLS methodology for the prediction of future seasons or new environments. Furthermore, the comparison between single univariate-trait (UT) versus MT for GBLUP and PLS gave an increase in prediction accuracy of MT-GBLUP versus UT-GBLUP, but not for MT-PLS versus UT-PLS.

Marker density and statistical model designs to increase accuracy of genomic selection for wool traits in Angora rabbits

The Angora rabbit, a well-known breed for fiber production, has been undergoing traditional breeding programs relying mainly on phenotypes. Genomic selection (GS) uses genomic information and promises to accelerate genetic gain. Practically, to implement GS in Angora rabbit breeding, it is necessary to evaluate different marker densities and GS models to develop suitable strategies for an optimized breeding pipeline. Considering a lack in microarray, low-coverage sequencing combined with genotype imputation was used to boost the number of SNPs across the rabbit genome. Here, in a population of 629 Angora rabbits, a total of 18,577,154 high-quality SNPs were imputed (imputation accuracy above 98%) based on low-coverage sequencing of 3.84X genomic coverage, and wool traits and body weight were measured at 70, 140 and 210 days of age. From the original markers, 0.5K, 1K, 3K, 5K, 10K, 50K, 100K, 500K, 1M and 2M were randomly selected and evaluated, resulting in 50K markers as the baseline for the heritability estimation and genomic prediction. Comparing to the GS performance of single-trait models, the prediction accuracy of nearly all traits could be improved by multi-trait models, which might because multiple-trait models used information from genetically correlated traits. Furthermore, we observed high significant negative correlation between the increased prediction accuracy from single-trait to multiple-trait models and estimated heritability. The results indicated that low-heritability traits could borrow more information from correlated traits and hence achieve higher prediction accuracy. The research first reported heritability estimation in rabbits by using genome-wide markers, and provided 50K as an optimal marker density for further microarray design, genetic evaluation and genomic selection in Angora rabbits. We expect that the work could provide strategies for GS in early selection, and optimize breeding programs in rabbits.

Comparison of single-trait and multi-trait genomic predictions on agronomic and disease resistance traits in spring wheat

This study performed comprehensive analyses on the predictive abilities of single-trait and two multi-trait models in three populations. Our results demonstrated the superiority of multi-traits over single-trait models across seven agronomic and four to seven disease resistance traits of different genetic architecture. The predictive ability of multi-trait and single-trait prediction models has not been investigated on diverse traits evaluated under organic and conventional management systems. Here, we compared the predictive abilities of 25% of a testing set that has not been evaluated for a single trait (ST), not evaluated for multi-traits (MT1), and evaluated for some traits but not others (MT2) in three spring wheat populations genotyped either with the wheat 90K single nucleotide polymorphisms array or DArTseq. Analyses were performed on seven agronomic traits evaluated under conventional and organic management systems, four to seven disease resistance traits, and all agronomic and disease resistance traits simultaneously. The average prediction accuracies of the ST, MT1, and MT2 models varied from 0.03 to 0.78 (mean 0.41), from 0.05 to 0.82 (mean 0.47), and from 0.05 to 0.92 (mean 0.67), respectively. The predictive ability of the MT2 model was significantly greater than the ST model in all traits and populations except common bunt with the MT1 model being intermediate between them. The MT2 model increased prediction accuracies over the ST and MT1 models in all traits by 9.0-82.4% (mean 37.3%) and 2.9-82.5% (mean 25.7%), respectively, except common bunt that showed up to 7.7% smaller accuracies in two populations. A joint analysis of all agronomic and disease resistance traits further improved accuracies within the MT1 and MT2 models on average by 21.4% and 17.4%, respectively, as compared to either the agronomic or disease resistance traits, demonstrating the high potential of the multi-traits models in improving prediction accuracies.

Genomic Prediction Strategies for Dry-Down-Related Traits in Maize

For efficient mechanical harvesting, low grain moisture content at harvest time is essential. Dry-down rate (DR), which refers to the reduction in grain moisture content after the plants enter physiological maturity, is one of the main factors affecting the amount of moisture in the kernels. Dry-down rate is estimated using kernel moisture content at physiological maturity and at harvest time; however, measuring kernel water content at physiological maturity, which is sometimes referred as kernel water content at black layer formation (BWC), is time-consuming and resource-demanding. Therefore, inferring BWC from other correlated and easier to measure traits could improve the efficiency of breeding efforts for dry-down-related traits. In this study, multi-trait genomic prediction models were used to estimate genetic correlations between BWC and water content at harvest time (HWC) and flowering time (FT). The results show there is moderate-to-high genetic correlation between the traits (0.24-0.66), which supports the use of multi-trait genomic prediction models. To investigate genomic prediction strategies, several cross-validation scenarios representing possible implementations of genomic prediction were evaluated. The results indicate that, in most scenarios, the use of multi-trait genomic prediction models substantially increases prediction accuracy. Furthermore, the inclusion of historical records for correlated traits can improve prediction accuracy, even when the target trait is not measured on all the plots in the training set.

Bayesian multitrait kernel methods improve multienvironment genome-based prediction

When multitrait data are available, the preferred models are those that are able to account for correlations between phenotypic traits because when the degree of correlation is moderate or large, this increases the genomic prediction accuracy. For this reason, in this article, we explore Bayesian multitrait kernel methods for genomic prediction and we illustrate the power of these models with three-real datasets. The kernels under study were the linear, Gaussian, polynomial, and sigmoid kernels; they were compared with the conventional Ridge regression and GBLUP multitrait models. The results show that, in general, the Gaussian kernel method outperformed conventional Bayesian Ridge and GBLUP multitrait linear models by 2.2–17.45% (datasets 1–3) in terms of prediction performance based on the mean square error of prediction. This improvement in terms of prediction performance of the Bayesian multitrait kernel method can be attributed to the fact that the proposed model is able to capture nonlinear patterns more efficiently than linear multitrait models. However, not all kernels perform well in the datasets used for evaluation, which is why more than one kernel should be evaluated to be able to choose the best kernel.

Genome-wide association mapping and genomic prediction of yield-related traits and starch pasting properties in cassava

GWAS identified eight yield-related, peak starch type of waxy and wild-type starch and 21 starch pasting property-related traits (QTLs). Prediction ability of eight GS models resulted in low to high predictability, depending on trait, heritability, and genetic architecture. Cassava is both a food and an industrial crop in Africa, South America, and Asia, but knowledge of the genes that control yield and starch pasting properties remains limited. We carried out a genome-wide association study to clarify the molecular mechanisms underlying these traits and to explore marker-based breeding approaches. We estimated the predictive ability of genomic selection (GS) using parametric, semi-parametric, and nonparametric GS models with a panel of 276 cassava genotypes from Thai Tapioca Development Institute, International Center for Tropical Agriculture, International Institute of Tropical Agriculture, and other breeding programs. The cassava panel was genotyped via genotyping-by-sequencing, and 89,934 single-nucleotide polymorphism (SNP) markers were identified. A total of 31 SNPs associated with yield, starch type, and starch properties traits were detected by the fixed and random model circulating probability unification (FarmCPU), Bayesian-information and linkage-disequilibrium iteratively nested keyway and compressed mixed linear model, respectively. GS models were developed, and forward predictabilities using all the prediction methods resulted in values of - 0.001-0.71 for the four yield-related traits and 0.33-0.82 for the seven starch pasting property traits. This study provides additional insight into the genetic architecture of these important traits for the development of markers that could be used in cassava breeding programs.

Can Cross-Country Genomic Predictions Be a Reasonable Strategy to Support Germplasm Exchange? - A Case Study With Hydrogen Cyanide in Cassava

Genomic prediction (GP) offers great opportunities for accelerated genetic gains by optimizing the breeding pipeline. One of the key factors to be considered is how the training populations (TP) are composed in terms of genetic improvement, kinship/origin, and their impacts on GP. Hydrogen cyanide content (HCN) is a determinant trait to guide cassava's products usage and processing. This work aimed to achieve the following objectives: (i) evaluate the feasibility of using cross-country (CC) GP between germplasm's of Embrapa Mandioca e Fruticultura (Embrapa, Brazil) and The International Institute of Tropical Agriculture (IITA, Nigeria) for HCN; (ii) provide an assessment of population structure for the joint dataset; (iii) estimate the genetic parameters based on single nucleotide polymorphisms (SNPs) and a haplotype-approach. Datasets of HCN from Embrapa and IITA breeding programs were analyzed, separately and jointly, with 1,230, 590, and 1,820 clones, respectively. After quality control, ∼14K SNPs were used for GP. The genomic estimated breeding values (GEBVs) were predicted based on SNP effects from analyses with TP composed of the following: (i) Embrapa genotypic and phenotypic data, (ii) IITA genotypic and phenotypic data, and (iii) the joint datasets. Comparisons on GEBVs' estimation were made considering the hypothetical situation of not having the phenotypic characterization for a set of clones for a certain research institute/country and might need to use the markers' effects that were trained with data from other research institutes/country's germplasm to estimate their clones' GEBV. Fixation index (FST) among the genetic groups identified within the joint dataset ranged from 0.002 to 0.091. The joint dataset provided an improved accuracy (0.8-0.85) compared to the prediction accuracy of either germplasm's sources individually (0.51-0.67). CC GP proved to have potential use under the present study's scenario, the correlation between GEBVs predicted with TP from Embrapa and IITA was 0.55 for Embrapa's germplasm, whereas for IITA's it was 0.1. This seems to be among the first attempts to evaluate the CC GP in plants. As such, a lot of useful new information was provided on the subject, which can guide new research on this very important and emerging field.

A population based expression atlas provides insights into disease resistance and other physiological traits in cassava (Manihot esculenta Crantz)

Cassava, a food security crop in Africa, is grown throughout the tropics and subtropics. Although cassava can provide high productivity in suboptimal conditions, the yield in Africa is substantially lower than in other geographies. The yield gap is attributable to many challenges faced by cassava in Africa, including susceptibility to diseases and poor soil conditions. In this study, we carried out 3’RNA sequencing on 150 accessions from the National Crops Resources Research Institute, Uganda for 5 tissue types, providing population-based transcriptomics resources to the research community in a web-based queryable cassava expression atlas. Differential expression and weighted gene co-expression network analysis were performed to detect 8820 significantly differentially expressed genes (DEGs), revealing similarity in expression patterns between tissue types and the clustering of detected DEGs into 18 gene modules. As a confirmation of data quality, differential expression and pathway analysis targeting cassava mosaic disease (CMD) identified 27 genes observed in the plant–pathogen interaction pathway, several previously identified CMD resistance genes, and two peroxidase family proteins different from the CMD2 gene. Present research work represents a novel resource towards understanding complex traits at expression and molecular levels for the development of resistant and high-yielding cassava varieties, as exemplified with CMD.

Genomic mating in outbred species: predicting cross usefulness with additive and total genetic covariance matrices

Diverse crops are both outbred and clonally propagated. Breeders typically use truncation selection of parents and invest significant time, land, and money evaluating the progeny of crosses to find exceptional genotypes. We developed and tested genomic mate selection criteria suitable for organisms of arbitrary homozygosity level where the full-sibling progeny are of direct interest as future parents and/or cultivars. We extended cross variance and covariance variance prediction to include dominance effects and predicted the multivariate selection index genetic variance of crosses based on haplotypes of proposed parents, marker effects, and recombination frequencies. We combined the predicted mean and variance into usefulness criteria for parent and variety development. We present an empirical study of cassava (Manihot esculenta), a staple tropical root crop. We assessed the potential to predict the multivariate genetic distribution (means, variances, and trait covariances) of 462 cassava families in terms of additive and total value using cross-validation. Most variance (89%) and covariance (70%) prediction accuracy estimates were greater than zero. The usefulness of crosses was accurately predicted with good correspondence between the predicted and the actual mean performance of family members breeders selected for advancement as new parents and candidate varieties. We also used a directional dominance model to quantify significant inbreeding depression for most traits. We predicted 47,083 possible crosses of 306 parents and contrasted them to those previously tested to show how mate selection can reveal the new potential within the germplasm. We enable breeders to consider the potential of crosses to produce future parents (progeny with top breeding values) and varieties (progeny with top own performance).

Accuracies of genomic predictions for disease resistance of striped catfish to Edwardsiella ictaluri using artificial intelligence algorithms

Assessments of genomic prediction accuracies using artificial intelligent (AI) algorithms (i.e., machine and deep learning methods) are currently not available or very limited in aquaculture species. The principal aim of this study was to examine the predictive performance of these new methods for disease resistance to Edwardsiella ictaluri in a population of striped catfish Pangasianodon hypophthalmus and to make comparisons with four common methods, i.e., pedigree-based best linear unbiased prediction (PBLUP), genomic-based best linear unbiased prediction (GBLUP), single-step GBLUP (ssGBLUP) and a nonlinear Bayesian approach (notably BayesR). Our analyses using machine learning (i.e., ML-KAML) and deep learning (i.e., DL-MLP and DL-CNN) together with the four common methods (PBLUP, GBLUP, ssGBLUP, and BayesR) were conducted for two main disease resistance traits (i.e., survival status coded as 0 and 1 and survival time, i.e., days that the animals were still alive after the challenge test) in a pedigree consisting of 560 individual animals (490 offspring and 70 parents) genotyped for 14,154 single nucleotide polymorphism (SNPs). The results using 6,470 SNPs after quality control showed that machine learning methods outperformed PBLUP, GBLUP, and ssGBLUP, with the increases in the prediction accuracies for both traits by 9.1–15.4%. However, the prediction accuracies obtained from machine learning methods were comparable to those estimated using BayesR. Imputation of missing genotypes using AlphaFamImpute increased the prediction accuracies by 5.3–19.2% in all the methods and data used. On the other hand, there were insignificant decreases (0.3–5.6%) in the prediction accuracies for both survival status and survival time when multivariate models were used in comparison to univariate analyses. Interestingly, the genomic prediction accuracies based on only highly significant SNPs (P < 0.00001, 318–400 SNPs for survival status and 1,362–1,589 SNPs for survival time) were somewhat lower (0.3–15.6%) than those obtained from the whole set of 6,470 SNPs. In most of our analyses, the accuracies of genomic prediction were somewhat higher for survival time than survival status (0/1 data). It is concluded that although there are prospects for the application of genomic selection to increase disease resistance to E. ictaluri in striped catfish breeding programs, further evaluation of these methods should be made in independent families/populations when more data are accumulated in future generations to avoid possible biases in the genetic parameters estimates and prediction accuracies for the disease-resistant traits studied in this population of striped catfish P. hypophthalmus.

RNA polymerase mapping in plants identifies intergenic regulatory elements enriched in causal variants

Control of gene expression is fundamental at every level of cell function. Promoter-proximal pausing and divergent transcription at promoters and enhancers, which are prominent features in animals, have only been studied in a handful of research experiments in plants. PRO-Seq analysis in cassava (Manihot esculenta) identified peaks of transcriptionally engaged RNA polymerase at both the 5' and 3' end of genes, consistent with paused or slowly moving Polymerase. In addition, we identified divergent transcription at intergenic sites. A full genome search for bi-directional transcription using an algorithm for enhancer detection developed in mammals (dREG) identified many intergenic regulatory element (IRE) candidates. These sites showed distinct patterns of methylation and nucleotide conservation based on genomic evolutionary rate profiling (GERP). SNPs within these IRE candidates explained significantly more variation in fitness and root composition than SNPs in chromosomal segments randomly ascertained from the same intergenic distribution, strongly suggesting a functional importance of these sites. Maize GRO-Seq data showed RNA polymerase occupancy at IREs consistent with patterns in cassava. Furthermore, these IREs in maize significantly overlapped with sites previously identified on the basis of open chromatin, histone marks, and methylation, and were enriched for reported eQTL. Our results suggest that bidirectional transcription can identify intergenic genomic regions in plants that play an important role in transcription regulation and whose identification has the potential to aid crop improvement.

Fifty years of a public cassava breeding program: evolution of breeding objectives, methods, and decision-making processes

This paper reviews and analyzes key features from cassava breeding at the International Center for Tropical Agriculture (CIAT) over 50 years and draws lessons for public breeding efforts broadly. The breeding team, jointly with national program partners and the private processing sector, defined breeding objectives and guiding business plans. These have evolved through the decades and currently focus on four global product profiles. The recurrent selection method also evolved and included innovations such as estimation of phenotypic breeding values, increasing the number of locations in the first stage of agronomic evaluations, gradual reduction of the duration of breeding cycles (including rapid cycling for high-heritability traits), the development of protocols for the induction of flowering, and the introduction of genome-wide predictions. The impact of cassava breeding depends significantly on the type of target markets. When roots are used for large processing facilities for starch, animal feeding or ethanol production (such as in SE Asia), the adoption of improved varieties is nearly universal and productivity at the regional scale increases significantly. When markets and relevant infrastructure are weak or considerable proportion of the production goes for local artisanal processing and on-farm consumption, the impact has been lower. The potential of novel breeding tools needs to be properly assessed for the most effective allocation of resources. Finally, a brief summary of challenges and opportunities for the future of cassava breeding is presented. The paper describes multiple ways that public and private sector breeding programs can learn from each other to optimize success.

Heterosis and Hybrid Crop Breeding: A Multidisciplinary Review

Although hybrid crop varieties are among the most popular agricultural innovations, the rationale for hybrid crop breeding is sometimes misunderstood. Hybrid breeding is slower and more resource-intensive than inbred breeding, but it allows systematic improvement of a population by recurrent selection and exploitation of heterosis simultaneously. Inbred parental lines can identically reproduce both themselves and their F1 progeny indefinitely, whereas outbred lines cannot, so uniform outbred lines must be bred indirectly through their inbred parents to harness heterosis. Heterosis is an expected consequence of whole-genome non-additive effects at the population level over evolutionary time. Understanding heterosis from the perspective of molecular genetic mechanisms alone may be elusive, because heterosis is likely an emergent property of populations. Hybrid breeding is a process of recurrent population improvement to maximize hybrid performance. Hybrid breeding is not maximization of heterosis per se, nor testing random combinations of individuals to find an exceptional hybrid, nor using heterosis in place of population improvement. Though there are methods to harness heterosis other than hybrid breeding, such as use of open-pollinated varieties or clonal propagation, they are not currently suitable for all crops or production environments. The use of genomic selection can decrease cycle time and costs in hybrid breeding, particularly by rapidly establishing heterotic pools, reducing testcrossing, and limiting the loss of genetic variance. Open questions in optimal use of genomic selection in hybrid crop breeding programs remain, such as how to choose founders of heterotic pools, the importance of dominance effects in genomic prediction, the necessary frequency of updating the training set with phenotypic information, and how to maintain genetic variance and prevent fixation of deleterious alleles.

Deep Learning for Predicting Complex Traits in Spring Wheat Breeding Program

Genomic selection (GS) is transforming the field of plant breeding and implementing models that improve prediction accuracy for complex traits is needed. Analytical methods for complex datasets traditionally used in other disciplines represent an opportunity for improving prediction accuracy in GS. Deep learning (DL) is a branch of machine learning (ML) which focuses on densely connected networks using artificial neural networks for training the models. The objective of this research was to evaluate the potential of DL models in the Washington State University spring wheat breeding program. We compared the performance of two DL algorithms, namely multilayer perceptron (MLP) and convolutional neural network (CNN), with ridge regression best linear unbiased predictor (rrBLUP), a commonly used GS model. The dataset consisted of 650 recombinant inbred lines (RILs) from a spring wheat nested association mapping (NAM) population planted from 2014-2016 growing seasons. We predicted five different quantitative traits with varying genetic architecture using cross-validations (CVs), independent validations, and different sets of SNP markers. Hyperparameters were optimized for DL models by lowering the root mean square in the training set, avoiding model overfitting using dropout and regularization. DL models gave 0 to 5% higher prediction accuracy than rrBLUP model under both cross and independent validations for all five traits used in this study. Furthermore, MLP produces 5% higher prediction accuracy than CNN for grain yield and grain protein content. Altogether, DL approaches obtained better prediction accuracy for each trait, and should be incorporated into a plant breeder's toolkit for use in large scale breeding programs.

Combining Genomic and Phenomic Information for Predicting Grain Protein Content and Grain Yield in Spring Wheat

Genomics and high throughput phenomics have the potential to revolutionize the field of wheat (Triticum aestivum L.) breeding. Genomic selection (GS) has been used for predicting various quantitative traits in wheat, especially grain yield. However, there are few GS studies for grain protein content (GPC), which is a crucial quality determinant. Incorporation of secondary correlated traits in GS models has been demonstrated to improve accuracy. The objectives of this research were to compare performance of single and multi-trait GS models for predicting GPC and grain yield in wheat and to identify optimal growth stages for collecting secondary traits. We used 650 recombinant inbred lines from a spring wheat nested association mapping (NAM) population. The population was phenotyped over 3 years (2014-2016), and spectral information was collected at heading and grain filling stages. The ability to predict GPC and grain yield was assessed using secondary traits, univariate, covariate, and multivariate GS models for within and across cycle predictions. Our results indicate that GS accuracy increased by an average of 12% for GPC and 20% for grain yield by including secondary traits in the models. Spectral information collected at heading was superior for predicting GPC, whereas grain yield was more accurately predicted during the grain filling stage. Green normalized difference vegetation index had the largest effect on the prediction of GPC either used individually or with multiple indices in the GS models. An increased prediction ability for GPC and grain yield with the inclusion of secondary traits demonstrates the potential to improve the genetic gain per unit time and cost in wheat breeding.

Genomic Prediction and Selection for Fruit Traits in Winter Squash

Improving fruit quality is an important but challenging breeding goal in winter squash. Squash breeding in general is resource-intensive, especially in terms of space, and the biology of squash makes it difficult to practice selection on both parents. These restrictions translate to smaller breeding populations and limited use of greenhouse generations, which in turn, limit genetic gain per breeding cycle and increases cycle length. Genomic selection is a promising technology for improving breeding efficiency; yet, few studies have explored its use in horticultural crops. We present results demonstrating the predictive ability of whole-genome models for fruit quality traits. Predictive abilities for quality traits were low to moderate, but sufficient for implementation. To test the use of genomic selection for improving fruit quality, we conducted three rounds of genomic recurrent selection in a butternut squash (Cucurbita moschata) population. Selections were based on a fruit quality index derived from a multi-trait genomic selection model. Remnant seed from selected populations was used to assess realized gain from selection. Analysis revealed significant improvement in fruit quality index value and changes in correlated traits. This study is one of the first empirical studies to evaluate gain from a multi-trait genomic selection model in a resource-limited horticultural crop.

Use of multiple traits genomic prediction, genotype by environment interactions and spatial effect to improve prediction accuracy in yield data

Genomic selection has been extensively implemented in plant breeding schemes. Genomic selection incorporates dense genome-wide markers to predict the breeding values for important traits based on information from genotype and phenotype records on traits of interest in a reference population. To date, most relevant investigations have been performed using single trait genomic prediction models (STGP). However, records for several traits at once are usually documented for breeding lines in commercial breeding programs. By incorporating benefits from genetic characterizations of correlated phenotypes, multiple trait genomic prediction (MTGP) may be a useful tool for improving prediction accuracy in genetic evaluations. The objective of this study was to test whether the use of MTGP and including proper modeling of spatial effects can improve the prediction accuracy of breeding values in commercial barley and wheat breeding lines. We genotyped 1,317 spring barley and 1,325 winter wheat lines from a commercial breeding program with the Illumina 9K barley and 15K wheat SNP-chip (respectively) and phenotyped them across multiple years and locations. Results showed that the MTGP approach increased correlations between future performance and estimated breeding value of yields by 7% in barley and by 57% in wheat relative to using the STGP approach for each trait individually. Analyses combining genomic data, pedigree information, and proper modeling of spatial effects further increased the prediction accuracy by 4% in barley and 3% in wheat relative to the model using genomic relationships only. The prediction accuracy for yield in wheat and barley yield trait breeding, were improved by combining MTGP and spatial effects in the model.

Genetic mapping and genomic selection for maize stalk strength

BACKGROUND

Maize is one of the most important staple crops and is widely grown throughout the world. Stalk lodging can cause enormous yield losses in maize production. However, rind penetrometer resistance (RPR), which is recognized as a reliable measurement to evaluate stalk strength, has been shown to be efficient and useful for improving stalk lodging-resistance. Linkage mapping is an acknowledged approach for exploring the genetic architecture of target traits. In addition, genomic selection (GS) using whole genome markers enhances selection efficiency for genetically complex traits. In the present study, two recombinant inbred line (RIL) populations were utilized to dissect the genetic basis of RPR, which was evaluated in seven growth stages.

RESULTS

The optimal stages to measure stalk strength are the silking phase and stages after silking. A total of 66 and 45 quantitative trait loci (QTL) were identified in each RIL population. Several potential candidate genes were predicted according to the maize gene annotation database and were closely associated with the biosynthesis of cell wall components. Moreover, analysis of gene ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway further indicated that genes related to cell wall formation were involved in the determination of RPR. In addition, a multivariate model of genomic selection efficiently improved the prediction accuracy relative to a univariate model and a model considering RPR-relevant loci as fixed effects.

CONCLUSIONS

The genetic architecture of RPR is highly genetically complex. Multiple minor effect QTL are jointly involved in controlling phenotypic variation in RPR. Several pleiotropic QTL identified in multiple stages may contain reliable genes and can be used to develop functional markers for improving the selection efficiency of stalk strength. The application of genomic selection to RPR may be a promising approach to accelerate breeding process for improving stalk strength and enhancing lodging-resistance.

Novel Bayesian Networks for Genomic Prediction of Developmental Traits in Biomass Sorghum

The ability to connect genetic information between traits over time allow Bayesian networks to offer a powerful probabilistic framework to construct genomic prediction models. In this study, we phenotyped a diversity panel of 869 biomass sorghum (Sorghum bicolor (L.) Moench) lines, which had been genotyped with 100,435 SNP markers, for plant height (PH) with biweekly measurements from 30 to 120 days after planting (DAP) and for end-of-season dry biomass yield (DBY) in four environments. We evaluated five genomic prediction models: Bayesian network (BN), Pleiotropic Bayesian network (PBN), Dynamic Bayesian network (DBN), multi-trait GBLUP (MTr-GBLUP), and multi-time GBLUP (MTi-GBLUP) models. In fivefold cross-validation, prediction accuracies ranged from 0.46 (PBN) to 0.49 (MTr-GBLUP) for DBY and from 0.47 (DBN, DAP120) to 0.75 (MTi-GBLUP, DAP60) for PH. Forward-chaining cross-validation further improved prediction accuracies of the DBN, MTi-GBLUP and MTr-GBLUP models for PH (training slice: 30-45 DAP) by 36.4-52.4% relative to the BN and PBN models. Coincidence indices (target: biomass, secondary: PH) and a coincidence index based on lines (PH time series) showed that the ranking of lines by PH changed minimally after 45 DAP. These results suggest a two-level indirect selection method for PH at harvest (first-level target trait) and DBY (second-level target trait) could be conducted earlier in the season based on ranking of lines by PH at 45 DAP (secondary trait). With the advance of high-throughput phenotyping technologies, our proposed two-level indirect selection framework could be valuable for enhancing genetic gain per unit of time when selecting on developmental traits.

Multi-trait random regression models increase genomic prediction accuracy for a temporal physiological trait derived from high-throughput phenotyping

Random regression models (RRM) are used extensively for genomic inference and prediction of time-valued traits in animal breeding, but only recently have been used in plant systems. High-throughput phenotyping (HTP) platforms provide a powerful means to collect high-dimensional phenotypes throughout the growing season for large populations. However, to date, selection of an appropriate statistical genomic framework to integrate multiple temporal traits for genomic prediction in plants remains unexplored. Here, we demonstrate the utility of a multi-trait RRM (MT-RRM) for genomic prediction of daily water usage (WU) in rice (Oryza sativa) through joint modeling with shoot biomass (projected shoot area, PSA). Three hundred and fifty-seven accessions were phenotyped daily for WU and PSA over 20 days using a greenhouse-based HTP platform. MT-RRMs that modeled additive genetic and permanent environmental effects for both traits using quadratic Legendre polynomials were used to assess genomic correlations between traits and genomic prediction for WU. Predictive abilities of the MT-RRMs were assessed using two cross-validation (CV) scenarios. The first scenario was designed to predict genetic values for WU at all time points for a set of accessions with unobserved WU. The second scenario was designed to forecast future genetic values for WU for a panel of known accessions with records for WU at earlier time periods. In each scenario we evaluated two MT-RRMs in which PSA records were absent or available for time points in the testing population. Weak to strong genomic correlations between WU and PSA were observed across the days of imaging (0.29-0.870.38-0.80). In both CV scenarios, MT-RRMs showed better predictive abilities compared to single-trait RRM, and prediction accuracies were greatly improved when PSA records were available for the testing population. In summary, these frameworks provide an effective approach to predict temporal physiological traits that are difficult or expensive to quantify in large populations.

Genetic Correlation, Genome-Wide Association and Genomic Prediction of Portable NIRS Predicted Carotenoids in Cassava Roots

Random forests (RF) was used to correlate spectral responses to known wet chemistry carotenoid concentrations including total carotenoid content (TCC), all-trans β-carotene (ATBC), violaxanthin (VIO), lutein (LUT), 15-cis beta-carotene (15CBC), 13-cis beta-carotene (13CBC), alpha-carotene (AC), 9-cis beta-carotene (9CBC), and phytoene (PHY) from laboratory analysis of 173 cassava root samples in Columbia. The cross-validated correlations between the actual and estimated carotenoid values using RF ranged from 0.62 in PHY to 0.97 in ATBC. The developed models were used to evaluate the carotenoids of 594 cassava clones with spectral information collected across three locations in a national breeding program (NRCRI, Umudike), Nigeria. Both populations contained cassava clones characterized as white and yellow. The NRCRI evaluated phenotypes were used to assess the genetic correlations, conduct genome-wide association studies (GWAS), and genomic predictions. Estimates of genetic correlation showed various levels of the relationship among the carotenoids. The associations between TCC and the individual carotenoids were all significant (P < 0.001) with high positive values (r > 0.75, except in LUT and PHY where r < 0.3). The GWAS revealed significant genomic regions on chromosomes 1, 2, 4, 13, 14, and 15 associated with variation in at least one of the carotenoids. One of the identified candidate genes, phytoene synthase (PSY) has been widely reported for variation in TCC in cassava. On average, genomic prediction accuracies from the single-trait genomic best linear unbiased prediction (GBLUP) and RF as well as from a multiple-trait GBLUP model ranged from ∼0.2 in LUT and PHY to 0.52 in TCC. The multiple-trait GBLUP model gave slightly higher accuracies than the single trait GBLUP and RF models. This study is one of the initial attempts in understanding the genetic basis of individual carotenoids and demonstrates the usefulness of NIRS in cassava improvement.

Historical Introgressions from a Wild Relative of Modern Cassava Improved Important Traits and May Be Under Balancing Selection

Introgression of alleles from wild relatives has often been adaptive in plant breeding. However, the significance of historical hybridization events in modern breeding is often not clear. Cassava (Manihot esculenta) is among the most important staple foods in the world, sustaining hundreds of millions of people in the tropics, especially in sub-Saharan Africa. Widespread genotyping makes cassava a model for clonally propagated root and tuber crops in the developing world, and provides an opportunity to study the modern benefits and consequences of historical introgression. We detected large introgressed Manihot glaziovii genome-segments in a collection of 2742 modern cassava landraces and elite germplasm, the legacy of a 1930s era breeding to combat disease epidemics. African landraces and improved varieties were, on average, 3.8% (max 13.6%) introgressed. Introgressions accounted for a significant (mean 20%, max 56%) portion of the heritability of tested traits. M. glaziovii alleles on the distal 10 Mb of chr. 1 increased dry matter and root number. On chr. 4, introgressions in a 20 Mb region improved harvest index and brown streak disease tolerance. We observed the introgression frequency on chr. 1 double over three cycles of selection, and that later stage trials selectively excluded homozygotes from consideration as varieties. This indicates a heterozygous advantage of introgressions. However, we also found that maintaining large recombination-suppressed introgressions in the heterozygous state allowed the accumulation of deleterious mutations. We conclude that targeted recombination of introgressions would increase the efficiency of cassava breeding by allowing simultaneous fixation of beneficial alleles and purging of genetic load.

A framework for genomics-informed ecophysiological modeling in plants

Dynamic process-based plant models capture complex physiological response across time, carrying the potential to extend simulations out to novel environments and lend mechanistic insight to observed phenotypes. Despite the translational opportunities for varietal crop improvement that could be unlocked by linking natural genetic variation to first principles-based modeling, these models are challenging to apply to large populations of related individuals. Here we use a combination of model development, experimental evaluation, and genomic prediction in Brassica rapa L. to set the stage for future large-scale process-based modeling of intraspecific variation. We develop a new canopy growth submodel for B. rapa within the process-based model Terrestrial Regional Ecosystem Exchange Simulator (TREES), test input parameters for feasibility of direct estimation with observed phenotypes across cultivated morphotypes and indirect estimation using genomic prediction on a recombinant inbred line population, and explore model performance on an in silico population under non-stressed and mild water-stressed conditions. We find evidence that the updated whole-plant model has the capacity to distill genotype by environment interaction (G×E) into tractable components. The framework presented offers a means to link genetic variation with environment-modulated plant response and serves as a stepping stone towards large-scale prediction of unphenotyped, genetically related individuals under untested environmental scenarios.

Multi-trait multi-environment models in the genetic selection of segregating soybean progeny

At present, single-trait best linear unbiased prediction (BLUP) is the standard method for genetic selection in soybean. However, when genetic selection is performed based on two or more genetically correlated traits and these are analyzed individually, selection bias may arise. Under these conditions, considering the correlation structure between the evaluated traits may provide more-accurate genetic estimates for the evaluated parameters, even under environmental influences. The present study was thus developed to examine the efficiency and applicability of multi-trait multi-environment (MTME) models by the residual maximum likelihood (REML/BLUP) and Bayesian approaches in the genetic selection of segregating soybean progeny. The study involved data pertaining to 203 soybean F2:4 progeny assessed in two environments for the following traits: number of days to maturity (DM), 100-seed weight (SW), and average seed yield per plot (SY). Variance components and genetic and non-genetic parameters were estimated via the REML/BLUP and Bayesian methods. The variance components estimated and the breeding values and genetic gains predicted with selection through the Bayesian procedure were similar to those obtained by REML/BLUP. The frequentist and Bayesian MTME models provided higher estimates of broad-sense heritability per plot (or heritability of total effects of progeny; [Formula: see text]) and mean accuracy of progeny than their respective single-trait versions. Bayesian analysis provided the credibility intervals for the estimates of [Formula: see text]. Therefore, MTME led to greater predicted gains from selection. On this basis, this procedure can be efficiently applied in the genetic selection of segregating soybean progeny.

Integrating genomic-enabled prediction and high-throughput phenotyping in breeding for climate-resilient bread wheat

Genomic selection and high-throughput phenotyping (HTP) are promising tools to accelerate breeding gains for high-yielding and climate-resilient wheat varieties. Hence, our objective was to evaluate them for predicting grain yield (GY) in drought-stressed (DS) and late-sown heat-stressed (HS) environments of the International maize and wheat improvement center's elite yield trial nurseries. We observed that the average genomic prediction accuracies using fivefold cross-validations were 0.50 and 0.51 in the DS and HS environments, respectively. However, when a different nursery/year was used to predict another nursery/year, the average genomic prediction accuracies in the DS and HS environments decreased to 0.18 and 0.23, respectively. While genomic predictions clearly outperformed pedigree-based predictions across nurseries, they were similar to pedigree-based predictions within nurseries due to small family sizes. In populations with some full-sibs in the training population, the genomic and pedigree-based prediction accuracies were on average 0.27 and 0.35 higher than the accuracies in populations with only one progeny per cross, indicating the importance of genetic relatedness between the training and validation populations for good predictions. We also evaluated the item-based collaborative filtering approach for multivariate prediction of GY using the green normalized difference vegetation index from HTP. This approach proved to be the best strategy for across-nursery predictions, with average accuracies of 0.56 and 0.62 in the DS and HS environments, respectively. We conclude that GY is a challenging trait for across-year predictions, but GS and HTP can be integrated in increasing the size of populations screened and evaluating unphenotyped large nurseries for stress-resilience within years.

Impact of residual covariance structures on genomic prediction ability in multi-environment trials

In plant breeding, one of the main purpose of multi-environment trial (MET) is to assess the intensity of genotype-by-environment (G×E) interactions in order to select high-performing lines of each environment. Most models to analyze such MET data consider only the additive genetic effects and the part of the non-additive genetic effects are confounded with the residual terms and this may lead to the non-negligible residual covariances between the same trait measured at multiple environments. In breeding programs it is also common to have the phenotype information from some environments available and values are missing in some other environments. In this study we focused on two problems: (1) to study the impact of different residual covariance structures on genomic prediction ability using different models to analyze MET data; (2) to compare the ability of different MET analysis models to predict the missing values in a single environment. Our results suggests that, it is important to consider the heterogeneous residual covariance structure for the MET analysis and multivariate mixed model seems to be especially suitable to predict the missing values in a single environment. We also present the prediction abilities based on Bayesian and frequentist approaches with different models using field data sets (maize and rice) having different levels of G×E interactions.