Wheat Dataset Research Articles

Wheat Ear Recognition Based on RetinaNet and Transfer Learning.

The number of wheat ears is an essential indicator for wheat production and yield estimation, but accurately obtaining wheat ears requires expensive manual cost and labor time. Meanwhile, the characteristics of wheat ears provide less information, and the color is consistent with the background, which can be challenging to obtain the number of wheat ears required. In this paper, the performance of Faster regions with convolutional neural networks (Faster R-CNN) and RetinaNet to predict the number of wheat ears for wheat at different growth stages under different conditions is investigated. The results show that using the Global WHEAT dataset for recognition, the RetinaNet method, and the Faster R-CNN method achieve an average accuracy of 0.82 and 0.72, with the RetinaNet method obtaining the highest recognition accuracy. Secondly, using the collected image data for recognition, the R2 of RetinaNet and Faster R-CNN after transfer learning is 0.9722 and 0.8702, respectively, indicating that the recognition accuracy of the RetinaNet method is higher on different data sets. We also tested wheat ears at both the filling and maturity stages; our proposed method has proven to be very robust (the R2 is above 90). This study provides technical support and a reference for automatic wheat ear recognition and yield estimation.

Adaptive Variable Re-weighting and Shrinking Approach for Variable Selection in Multivariate Calibration for Near-infrared Spectroscopy

Abstract With the generation of high-dimensional data from spectroscopic instruments, the role of variable selection in spectral modeling has become very important. This research proposes a new variable selection algorithm, named adaptive variable re-weighting and shrinking approach (AVRSA), based on model population analysis (MPA) and weighted bootstrap sampling (WBS). In this algorithm, WBS is used to generate sub-datasets for modeling in each iteration round, and the variable weight and space are updated by statistically evaluating the optimal sub-models. Unlike most other variable selection methods, the average prediction performance of the optimal sub-models by AVRSA must be preferable to that of the previous iteration. The best informative variables are obtained until no further optimal sub-models are generated. This method is checked on three near infrared (NIR) datasets. Three variable selection methods, including competitive adaptive reweighted sampling (CARS), MonteCarlo uninformative variable elimination (MC-UVE) and iteratively variable subset optimization (IVSO), are used for comparison. Compared with these variable selection algorithms, AVRSA selects the least informative variables, which is convenient for the development of portable instruments. Compared with the full-spectrum PLS model, the root mean square error of the validation set (RMSEP) of corn starch is decreased from 0.2614 to 0.1093, and the RMSEP of corn protein is decreased from 0.0977 to 0.0374. In addition, the RMSEP of the wheat dataset is decreased from 0.2585 to 0.2157, and the RMSEP of the wheat kernel dataset is decreased from 0.7816 to 0.6661. The results show that the proposed method is very efficient for the high-dimensional spectrum to find the best variables and improve the model's prediction performance.

Using noise reduction to enhance ranking based genomic selection

Genomic Selection (GS) is a breeding technique that utilizes whole genome markers to make trait predictions. The goal of GS is to identify the top candidates that have the most desirable trait values. Usually, GS has been formulated as a regression problem where the marker data is used to predict phenotypic values. However, since the end goal of GS is identification of top candidates, ranking the individuals makes far more sense. Creating accurate ranking models pose three fundamental challenges—presence of noise in phenotypic data, extremely high dimensional nature of the genotypic data and small sample size of the genomic datasets. To combat these challenges, we present a novel two phase approach to increase the noise tolerance of ranking based approaches. The proposed algorithm uses pruning to perform noise filtering and leverages biclustering to improve model generalization. This approach is evaluated on both pointwise and pairwise ranking algorithms. Previous work on Arabidopsis and CIMMYT wheat datasets yielded mean Normalized Discounted Cumulative Gain (NDCG) @10 scores of 0.883 and 0.748 respectively. The proposed approach outperforms these results on both of the datasets yielding ranking accuracies of 0.965 and 0.865 respectively.

Deep Segmentation of Point Clouds of Wheat.

The 3D analysis of plants has become increasingly effective in modeling the relative structure of organs and other traits of interest. In this paper, we introduce a novel pattern-based deep neural network, Pattern-Net, for segmentation of point clouds of wheat. This study is the first to segment the point clouds of wheat into defined organs and to analyse their traits directly in 3D space. Point clouds have no regular grid and thus their segmentation is challenging. Pattern-Net creates a dynamic link among neighbors to seek stable patterns from a 3D point set across several levels of abstraction using the K-nearest neighbor algorithm. To this end, different layers are connected to each other to create complex patterns from the simple ones, strengthen dynamic link propagation, alleviate the vanishing-gradient problem, encourage link reuse and substantially reduce the number of parameters. The proposed deep network is capable of analysing and decomposing unstructured complex point clouds into semantically meaningful parts. Experiments on a wheat dataset verify the effectiveness of our approach for segmentation of wheat in 3D space.

Optimal breeding-value prediction using a sparse selection index.

Genomic prediction uses DNA sequences and phenotypes to predict genetic values. In homogeneous populations, theory indicates that the accuracy of genomic prediction increases with sample size. However, differences in allele frequencies and linkage disequilibrium patterns can lead to heterogeneity in SNP effects. In this context, calibrating genomic predictions using a large, potentially heterogeneous, training data set may not lead to optimal prediction accuracy. Some studies tried to address this sample size/homogeneity trade-off using training set optimization algorithms; however, this approach assumes that a single training data set is optimum for all individuals in the prediction set. Here, we propose an approach that identifies, for each individual in the prediction set, a subset from the training data (i.e., a set of support points) from which predictions are derived. The methodology that we propose is a sparse selection index (SSI) that integrates selection index methodology with sparsity-inducing techniques commonly used for high-dimensional regression. The sparsity of the resulting index is controlled by a regularization parameter (λ); the G-Best Linear Unbiased Predictor (G-BLUP) (the prediction method most commonly used in plant and animal breeding) appears as a special case which happens when λ = 0. In this study, we present the methodology and demonstrate (using two wheat data sets with phenotypes collected in 10 different environments) that the SSI can achieve significant (anywhere between 5 and 10%) gains in prediction accuracy relative to the G-BLUP.

Application of multi-trait Bayesian decision theory for parental genomic selection.

In all breeding programs, the decision about which individuals to select and intermate to form the next selection cycle is crucial. The improvement of genetic stocks requires considering multiple traits simultaneously, given that economic value and net genetic merits depend on many traits; therefore, with the advance of computational and statistical tools and genomic selection (GS), researchers are focusing on multi-trait selection. Selection of the best individuals is difficult, especially in traits that are antagonistically correlated, where improvement in one trait might imply a reduction in other(s). There are approaches that facilitate multi-trait selection, and recently a Bayesian decision theory (BDT) has been proposed. Parental selection using BDT has the potential to be effective in multi-trait selection given that it summarizes all relevant quantitative genetic concepts such as heritability, response to selection and the structure of dependence between traits (correlation). In this study, we applied BDT to provide a treatment for the complexity of multi-trait parental selection using three multivariate loss functions (LF), Kullback–Leibler (KL), Energy Score, and Multivariate Asymmetric Loss (MALF), to select the best-performing parents for the next breeding cycle in two extensive real wheat data sets. Results show that the high ranking lines in genomic estimated breeding value (GEBV) for certain traits did not always have low values for the posterior expected loss (PEL). For both data sets, the KL LF gave similar importance to all traits including grain yield. In contrast, the Energy Score and MALF gave a better performance in three of four traits that were different than grain yield. The BDT approach should help breeders to decide based not only on the GEBV per se of the parent to be selected, but also on the level of uncertainty according to the Bayesian paradigm.

Phenotype Prediction Under Epistasis.

Reliable methods of phenotype prediction from genomic data play an increasingly important role in many areas of plant and animal breeding. Thus, developing methods that enhance prediction accuracy is of major interest. Here, we provide three methods for this purpose: (1) Genomic Best Linear Unbiased Prediction (GBLUP) as a model just accounting for additive SNP effects; (2) Epistatic Random Regression BLUP (ERRBLUP) as a full epistatic model which incorporates all pairwise SNP interactions, and (3) selective Epistatic Random Regression BLUP (sERRBLUP) as an epistatic model which incorporates a subset of pairwise SNP interactions selected based on their absolute effect sizes or the effect variances, which is computed based on solutions from the ERRBLUP model. We compared the predictive ability obtained from GBLUP, ERRBLUP, and sERRBLUP with genotypes from a publicly available wheat dataset and respective simulated phenotypes. Results showed that sERRBLUP provides a substantial increase in prediction accuracy compared to the other methods when the optimal proportion of SNP interactions is kept in the model, especially when an optimal proportion of SNP interactions is selected based on the SNP interaction effect sizes. All methods described here are implemented in the R-package EpiGP, which is able to process large-scale genomic data in a computationally efficient way.

Automatic Microplot Localization Using UAV Images and a Hierarchical Image-Based Optimization Method.

To develop new crop varieties and monitor plant growth, health, and traits, automated analysis of aerial crop images is an attractive alternative to time-consuming manual inspection. To perform per-microplot phenotypic analysis, localizing and detecting individual microplots in an orthomosaic image of a field are major steps. Our algorithm uses an automatic initialization of the known field layout over the orthomosaic images in roughly the right position. Since the orthomosaic images are stitched from a large number of smaller images, there can be distortion causing microplot rows not to be entirely straight and the automatic initialization to not correctly position every microplot. To overcome this, we have developed a three-level hierarchical optimization method. First, the initial bounding box position is optimized using an objective function that maximizes the level of vegetation inside the area. Then, columns of microplots are repositioned, constrained by their expected spacing. Finally, the position of microplots is adjusted individually using an objective function that simultaneously maximizes the area of the microplot overlapping vegetation, minimizes spacing variance between microplots, and maximizes each microplot's alignment relative to other microplots in the same row and column. The orthomosaics used in this study were obtained from multiple dates of canola and wheat breeding trials. The algorithm was able to detect 99.7% of microplots for canola and 99% for wheat. The automatically segmented microplots were compared to ground truth segmentations, resulting in an average DSC of 91.2% and 89.6% across all microplots and orthomosaics in the canola and wheat datasets.

WheatGmap: a comprehensive platform for wheat gene mapping and genomic studies

WheatGmap: a comprehensive platform for wheat gene mapping and genomic studies

Comparing the Potential of Marker-Assisted Selection and Genomic Prediction for Improving Rust Resistance in Hybrid Wheat.

Improving leaf rust and stripe rust resistance is a central goal in wheat breeding. The objectives of this study were to (1) elucidate the genetic basis of leaf rust and stripe rust resistance in a hybrid wheat population, (2) compare the findings using a previously published hybrid wheat data set, and (3) contrast the prediction accuracy with those of genome-wide prediction. The hybrid wheat population included 1,744 single crosses from 236 parental lines. The genotypes were fingerprinted using a 15k SNP array and evaluated for leaf rust and stripe rust resistance in multi-location field trials. We observed a high congruency of putative quantitative trait loci (QTL) for leaf rust resistance between both populations. This was not the case for stripe rust resistance. Accordingly, prediction accuracy of the detected QTL was moderate for leaf rust but low for stripe rust resistance. Genome-wide selection increased the prediction accuracy slightly for stripe rust albeit at a low level but not for leaf rust. Thus, our findings suggest that marker-assisted selection seems to be a robust and efficient tool to improve leaf rust resistance in European wheat hybrids.

Approximate Genome-Based Kernel Models for Large Data Sets Including Main Effects and Interactions.

The rapid development of molecular markers and sequencing technologies has made it possible to use genomic prediction (GP) and selection (GS) in animal and plant breeding. However, when the number of observations (n) is large (thousands or millions), computational difficulties when handling these large genomic kernel relationship matrices (inverting and decomposing) increase exponentially. This problem increases when genomic × environment interaction and multi-trait kernels are included in the model. In this research we propose selecting a small number of lines m(m < n) for constructing an approximate kernel of lower rank than the original and thus exponentially decreasing the required computing time. First, we describe the full genomic method for single environment (FGSE) with a covariance matrix (kernel) including all n lines. Second, we select m lines and approximate the original kernel for the single environment model (APSE). Similarly, but including main effects and G × E, we explain a full genomic method with genotype × environment model (FGGE), and including m lines, we approximated the kernel method with G × E (APGE). We applied the proposed method to two different wheat data sets of different sizes (n) using the standard linear kernel Genomic Best Linear Unbiased Predictor (GBLUP) and also using eigen value decomposition. In both data sets, we compared the prediction performance and computing time for FGSE versus APSE; we also compared FGGE versus APGE. Results showed a competitive prediction performance of the approximated methods with a significant reduction in computing time. Genomic prediction accuracy depends on the decay of the eigenvalues (amount of variance information loss) of the original kernel as well as on the size of the selected lines m.

Yield reduction under climate warming varies among wheat cultivars in South Africa

Understanding extreme weather impacts on staple crops such as wheat is vital for creating adaptation strategies and increasing food security, especially in dryland cropping systems across Southern Africa. This study analyses heat impacts on wheat using daily weather information and a dryland wheat dataset for 71 cultivars across 17 locations in South Africa from 1998 to 2014. We estimate temperature impacts on yields in extensive regression models, finding that extreme heat drives wheat yield losses, with an additional 24 h of exposure to temperatures above 30 °C associated with a 12.5% yield reduction. Results from a uniform warming scenario of +1 °C show an average wheat yield reduction of 8.5%, which increases to 18.4% and 28.5% under +2 and +3 °C scenarios. We also find evidence of differences in heat effects across cultivars, which suggests warming impacts may be reduced through the sharing of gene pools amongst wheat breeding programs.

Building a successful international research community through data sharing: The case of the Wheat Information System (WheatIS)

The International Wheat Information System (WheatIS) Expert Working Group (EWG) was initiated in 2012 under the Wheat Initiative with a broad range of contributing organizations. The mission of the WheatIS EWG was to create an informational infrastructure, establish data standards, and build a single portal that allows search, retrieval, and display of globally distributed wheat data sets that are indexed in standard data formats at servers around the world. The web portal at WheatIS.org was released publicly in 2015, and by 2020, it expanded to 8 geographically-distributed nodes and around 20 organizations under its umbrella. In this paper, we present our experience, the challenges we faced, and the answer we brought for establishing an international research community to build an informational infrastructure. Our hope is that our experience with building wheatis.org will guide current and future research communities to facilitate institutional and international challenges to create global tools and resources to help their respective scientific communities.

Building a successful international research community through data sharing: The case of the Wheat Information System (WheatIS).

The International Wheat Information System (WheatIS) Expert Working Group (EWG) was initiated in 2012 under the Wheat Initiative with a broad range of contributing organizations. The mission of the WheatIS EWG was to create an informational infrastructure, establish data standards, and build a single portal that allows search, retrieval, and display of globally distributed wheat data sets that are indexed in standard data formats at servers around the world. The web portal at WheatIS.org was released publicly in 2015, and by 2020, it expanded to 8 geographically-distributed nodes and around 20 organizations under its umbrella. In this paper, we present our experience, the challenges we faced, and the answer we brought for establishing an international research community to build an informational infrastructure. Our hope is that our experience with building wheatis.org will guide current and future research communities to facilitate institutional and international challenges to create global tools and resources to help their respective scientific communities.

Identification and characterization of long non-coding RNAs regulating resistant starch biosynthesis in bread wheat (Triticum aestivum L.)

Resistant starch (RS) also known as healthy starch has shown several health benefits. Enhancing the RS through genetic modification approaches has huge commercial importance. Regulatory RNA like long non-coding RNA (lncRNA) plays an important role in gene regulation. In this study, we mined 63 transcriptome datasets of wheat belonging to 35 genotypes representing two seed developmental stages. Contrasting expression of a subset oflncRNAs in RS mutant lines compared to parent wheat variety ‘C 306’ signifies their probable role in RS biosynthesis. Further, lncRNA- TCONS_00130663 showed strong positive correlation (r2 = 1) with LYPL gene and strong negative correlation with SBEIIb (r2 = −0.94). We found TCONS_00130663 as positive regulator of LYPL gene through interaction with miR1128. Based on relative expression, in silico interaction and DSC analysis we hypothesize the dual role of TCONS_00130663 in RS type 2 and type 5. The study provides a useful resource for functional mechanism of lncRNAs.

Inferring trait-specific similarity among individuals from molecular markers and phenotypes with Bayesian regression

Modeling covariance structure based on genetic similarity between pairs of relatives plays an important role in evolutionary, quantitative and statistical genetics. Historically, genetic similarity between individuals has been quantified from pedigrees via the probability that randomly chosen homologous alleles between individuals are identical by descent (IBD). At present, however, many genetic analyses rely on molecular markers, with realized measures of genomic similarity replacing IBD-based expected similarities. Animal and plant breeders, for example, now employ marker-based genomic relationship matrices between individuals in prediction models and in estimation of genome-based heritability coefficients. Phenotypes convey information about genetic similarity as well. For instance, if phenotypic values are at least partially the result of the action of quantitative trait loci, one would expect the former to inform about the latter, as in genome-wide association studies. Statistically, a non-trivial conditional distribution of unknown genetic similarities, given phenotypes, is to be expected. A Bayesian formalism is presented here that applies to whole-genome regression methods where some genetic similarity matrix, e.g., a genomic relationship matrix, can be defined. Our Bayesian approach, based on phenotypes and markers, converts prior (markers only) expected similarity into trait-specific posterior similarity. A simulation illustrates situations under which effective Bayesian learning from phenotypes occurs. Pinus and wheat data sets were used to demonstrate applicability of the concept in practice. The methodology applies to a wide class of Bayesian linear regression models, it extends to the multiple-trait domain, and can also be used to develop phenotype-guided similarity kernels in prediction problems.

Deep Kernel and Deep Learning for Genome-Based Prediction of Single Traits in Multienvironment Breeding Trials.

Deep learning (DL) is a promising method for genomic-enabled prediction. However, the implementation of DL is difficult because many hyperparameters (number of hidden layers, number of neurons, learning rate, number of epochs, batch size, etc.) need to be tuned. For this reason, deep kernel methods, which only require defining the number of layers, may be an attractive alternative. Deep kernel methods emulate DL models with a large number of neurons, but are defined by relatively easily computed covariance matrices. In this research, we compared the genome-based prediction of DL to a deep kernel (arc-cosine kernel, AK), to the commonly used non-additive Gaussian kernel (GK), as well as to the conventional additive genomic best linear unbiased predictor (GBLUP/GB). We used two real wheat data sets for benchmarking these methods. On average, AK and GK outperformed DL and GB. The gain in terms of prediction performance of AK and GK over DL and GB was not large, but AK and GK have the advantage that only one parameter, the number of layers (AK) or the bandwidth parameter (GK), has to be tuned in each method. Furthermore, although AK and GK had similar performance, deep kernel AK is easier to implement than GK, since the parameter “number of layers” is more easily determined than the bandwidth parameter of GK. Comparing AK and DL for the data set of year 2015–2016, the difference in performance of the two methods was bigger, with AK predicting much better than DL. On this data, the optimization of the hyperparameters for DL was difficult and the finally used parameters may have been suboptimal. Our results suggest that AK is a good alternative to DL with the advantage that practically no tuning process is required.

A Bayesian Genomic Multi-output Regressor Stacking Model for Predicting Multi-trait Multi-environment Plant Breeding Data.

In this paper we propose a Bayesian multi-output regressor stacking (BMORS) model that is a generalization of the multi-trait regressor stacking method. The proposed BMORS model consists of two stages: in the first stage, a univariate genomic best linear unbiased prediction (GBLUP including genotype × environment interaction GE) model is implemented for each of the traits under study; then the predictions of all traits are included as covariates in the second stage, by implementing a Ridge regression model. The main objectives of this research were to study alternative models to the existing multi-trait multi-environment (BMTME) model with respect to (1) genomic-enabled prediction accuracy, and (2) potential advantages in terms of computing resources and implementation. We compared the predictions of the BMORS model to those of the univariate GBLUP model using 7 maize and wheat datasets. We found that the proposed BMORS produced similar predictions to the univariate GBLUP model and to the BMTME model in terms of prediction accuracy; however, the best predictions were obtained under the BMTME model. In terms of computing resources, we found that the BMORS is at least 9 times faster than the BMTME method. Based on our empirical findings, the proposed BMORS model is an alternative for predicting multi-trait and multi-environment data, which are very common in genomic-enabled prediction in plant and animal breeding programs.

Validation of Candidate Gene-Based Markers and Identification of Novel Loci for Thousand-Grain Weight in Spring Bread Wheat.

Increased thousand-grain weight (TGW) is an important breeding target for indirectly improving grain yield (GY). Fourteen reported candidate genes known to enhance TGW were evaluated in two independent and existing datasets of wheat at CIMMYT, the Elite Yield Trial (EYT) from 2015 to 2016 (EYT2015-16) and the Wheat Association Mapping Initiative (WAMI) panel, to study their allele effects on TGW and to verify their suitability for marker-assisted selection. Of these, significant associations were detected for only one gene (TaGs3-D1) in the EYT2015-16 and two genes (TaTGW6 and TaSus1) in WAMI. The reported favorable alleles of TaGs3-D1 and TaTGW6 genes decreased TGW in the datasets. A haplotype-based genome wide association study was implemented to identify the genetic determinants of TGW on a large set of CIMMYT germplasm (4,302 lines comprising five EYTs), which identified 15 haplotype blocks to be significantly associated with TGW. Four of them, identified on chromosomes 4A, 6A, and 7A, were associated with TGW in at least three EYTs. The locus on chromosome 6A (Hap-6A-13) had the largest effect on TGW and additionally GY with increases of up to 2.60 g and 258 kg/ha, respectively. Discovery of novel TGW loci described in our study expands the opportunities for developing diagnostic markers and for multi-gene pyramiding to derive new allele combinations for enhanced TGW and GY in CIMMYT wheat.

AquaCropR: Crop Growth Model for R

The Food and Agriculture Organization (FAO) AquaCrop model, run either via a standalone graphical user interface (GUI) or via a matlab application programming interface (API) (AquaCrop-OS), has been successfully tested on many crop species and under multiple scenarios. However, with these current versions, it is difficult for users to adapt formulae, add functionality or incorporate the model into other applications such as decision support tools. Here, we report on the release of a version of AquaCrop written in R. Performance of the model was tested using published datasets of wheat (Triticum aestivum L.) and maize (Zea mays L.), comparing output from AquaCropR with these other versions of AquaCrop. Our goal in developing this version was to widen the use and improvement of AquaCrop through open access.