New Deep Learning Genomic-Based Prediction Model for Multiple Traits with Binary, Ordinal, and Continuous Phenotypes

Review of applications of artificial intelligence (AI) methods in crop research

Sophisticated and modern crop improvement techniques can bridge the gap for feeding the ever-increasing population. Artificial intelligence (AI) refers to the simulation of human intelligence in machines, which refers to the application of computational algorithms, machine learning (ML) and deep learning (DL) techniques. This is aimed to generalise patterns and relationships from historical data, employing various mathematical optimisation techniques thus making prediction models for facilitating selection of superior genotypes. These techniques are less resource intensive and can solve the problem based on the analysis of large-scale phenotypic datasets. ML for genomic selection (GS) uses high-throughput genotyping technologies to gather genetic information on a large number of markers across the genome. The prediction of GS models is based on the mathematical relation between genotypic and phenotypic data from the training population. ML techniques have emerged as powerful tools for genome editing through analysing large-scale genomic data and facilitating the development of accurate prediction models. Precise phenotyping is a prerequisite to advance crop breeding for solving agricultural production-related issues. ML algorithms can solve this problem through generating predictive models, based on the analysis of large-scale phenotypic datasets. DL models also have the potential reliability of precise phenotyping. This review provides a comprehensive overview on various ML and DL models, their applications, potential to enhance the efficiency, specificity and safety towards advanced crop improvement protocols such as genomic selection, genome editing, along with phenotypic prediction to promote accelerated breeding.

Application of deep learning with bivariate models for genomic prediction of sow lifetime productivity-related traits

OBJECTIVE

Pig breeders cannot obtain phenotypic information at the time of selection for sow lifetime productivity (SLP). They would benefit from obtaining genetic information of candidate sows. Genomic data interpreted using deep learning (DL) techniques could contribute to the genetic improvement of SLP to maximize farm profitability because DL models capture nonlinear genetic effects such as dominance and epistasis more efficiently than conventional genomic prediction methods based on linear models. This study aimed to investigate the usefulness of DL for the genomic prediction of two SLP-related traits; lifetime number of litters (LNL) and lifetime pig production (LPP).

METHODS

Two bivariate DL models, convolutional neural network (CNN) and local convolutional neural network (LCNN), were compared with conventional bivariate linear models (i.e., genomic best linear unbiased prediction, Bayesian ridge regression, Bayes A, and Bayes B). Phenotype and pedigree data were collected from 40,011 sows that had husbandry records. Among these, 3,652 pigs were genotyped using the PorcineSNP60K BeadChip.

RESULTS

The best predictive correlation for LNL was obtained with CNN (0.28), followed by LCNN (0.26) and conventional linear models (approximately 0.21). For LPP, the best predictive correlation was also obtained with CNN (0.29), followed by LCNN (0.27) and conventional linear models (approximately 0.25). A similar trend was observed with the mean squared error of prediction for the SLP traits.

CONCLUSION

This study provides an example of a CNN that can outperform against the linear model-based genomic prediction approaches when the nonlinear interaction components are important because LNL and LPP exhibited strong epistatic interaction components. Additionally, our results suggest that applying bivariate DL models could also contribute to the prediction accuracy by utilizing the genetic correlation between LNL and LPP.

Climate change impacts on crop breeding: Targeting interacting biotic and abiotic stresses for wheat improvement

Wheat (Triticum aestivum L.) as a staple crop is closely interwoven into the development of modern society. Its influence on culture and economic development is global. Recent instability in wheat markets has demonstrated its importance in guaranteeing food security across national borders. Climate change threatens food security as it interacts with a multitude of factors impacting wheat production. The challenge needs to be addressed with a multidisciplinary perspective delivered across research, private, and government sectors. Many experimental studies have identified the major biotic and abiotic stresses impacting wheat production, but fewer have addressed the combinations of stresses that occur simultaneously or sequentially during the wheat growth cycle. Here, we argue that biotic and abiotic stress interactions, and the genetics and genomics underlying them, have been insufficiently addressed by the crop science community. We propose this as a reason for the limited transfer of practical and feasible climate adaptation knowledge from research projects into routine farming practice. To address this gap, we propose that novel methodology integration can align large volumes of data available from crop breeding programs with increasingly cheaper omics tools to predict wheat performance under different climate change scenarios. Underlying this is our proposal that breeders design and deliver future wheat ideotypes based on new or enhanced understanding of the genetic and physiological processes that are triggered when wheat is subjected to combinations of stresses. By defining this to a trait and/or genetic level, new insights can be made for yield improvement under future climate conditions.

Genomic prediction using machine learning: a comparison of the performance of regularized regression, ensemble, instance-based and deep learning methods on synthetic and empirical data

BACKGROUND

The accurate prediction of genomic breeding values is central to genomic selection in both plant and animal breeding studies. Genomic prediction involves the use of thousands of molecular markers spanning the entire genome and therefore requires methods able to efficiently handle high dimensional data. Not surprisingly, machine learning methods are becoming widely advocated for and used in genomic prediction studies. These methods encompass different groups of supervised and unsupervised learning methods. Although several studies have compared the predictive performances of individual methods, studies comparing the predictive performance of different groups of methods are rare. However, such studies are crucial for identifying (i) groups of methods with superior genomic predictive performance and assessing (ii) the merits and demerits of such groups of methods relative to each other and to the established classical methods. Here, we comparatively evaluate the genomic predictive performance and informally assess the computational cost of several groups of supervised machine learning methods, specifically, regularized regression methods, deep, ensemble and instance-based learning algorithms, using one simulated animal breeding dataset and three empirical maize breeding datasets obtained from a commercial breeding program.

RESULTS

Our results show that the relative predictive performance and computational expense of the groups of machine learning methods depend upon both the data and target traits and that for classical regularized methods, increasing model complexity can incur huge computational costs but does not necessarily always improve predictive accuracy. Thus, despite their greater complexity and computational burden, neither the adaptive nor the group regularized methods clearly improved upon the results of their simple regularized counterparts. This rules out selection of one procedure among machine learning methods for routine use in genomic prediction. The results also show that, because of their competitive predictive performance, computational efficiency, simplicity and therefore relatively few tuning parameters, the classical linear mixed model and regularized regression methods are likely to remain strong contenders for genomic prediction.

CONCLUSIONS

The dependence of predictive performance and computational burden on target datasets and traits call for increasing investments in enhancing the computational efficiency of machine learning algorithms and computing resources.

Multimodal deep learning methods enhance genomic prediction of wheat breeding

While several statistical machine learning methods have been developed and studied for assessing the genomic prediction (GP) accuracy of unobserved phenotypes in plant breeding research, few methods have linked genomics and phenomics (imaging). Deep learning (DL) neural networks have been developed to increase the GP accuracy of unobserved phenotypes while simultaneously accounting for the complexity of genotype-environment interaction (GE); however, unlike conventional GP models, DL has not been investigated for when genomics is linked with phenomics. In this study we used 2 wheat data sets (DS1 and DS2) to compare a novel DL method with conventional GP models. Models fitted for DS1 were GBLUP, gradient boosting machine (GBM), support vector regression (SVR) and the DL method. Results indicated that for 1 year, DL provided better GP accuracy than results obtained by the other models. However, GP accuracy obtained for other years indicated that the GBLUP model was slightly superior to the DL. DS2 is comprised only of genomic data from wheat lines tested for 3 years, 2 environments (drought and irrigated) and 2-4 traits. DS2 results showed that when predicting the irrigated environment with the drought environment, DL had higher accuracy than the GBLUP model in all analyzed traits and years. When predicting drought environment with information on the irrigated environment, the DL model and GBLUP model had similar accuracy. The DL method used in this study is novel and presents a strong degree of generalization as several modules can potentially be incorporated and concatenated to produce an output for a multi-input data structure.

Crop genomic selection with deep learning and environmental data: A survey

Machine learning techniques for crop genomic selections, especially for single-environment plants, are well-developed. These machine learning models, which use dense genome-wide markers to predict phenotype, routinely perform well on single-environment datasets, especially for complex traits affected by multiple markers. On the other hand, machine learning models for predicting crop phenotype, especially deep learning models, using datasets that span different environmental conditions, have only recently emerged. Models that can accept heterogeneous data sources, such as temperature, soil conditions and precipitation, are natural choices for modeling GxE in multi-environment prediction. Here, we review emerging deep learning techniques that incorporate environmental data directly into genomic selection models.

Modeling genotype × environment interaction for single and multitrait genomic prediction in potato (Solanum tuberosum L.)

In this study, we extend research on genomic prediction (GP) to polysomic polyploid plant species with the main objective to investigate single-trait (ST) and multitrait (MT) multienvironment (ME) models using field trial data from 3 locations in Sweden [Helgegården (HEL), Mosslunda (MOS), Umeå (UM)] over 2 years (2020, 2021) of 253 potato cultivars and breeding clones for 5 tuber weight traits and 2 tuber flesh quality characteristics. This research investigated the GP of 4 genome-based prediction models with genotype × environment interactions (GEs): (1) ST reaction norm model (M1), (2) ST model considering covariances between environments (M2), (3) ST M2 extended to include a random vector that utilizes the environmental covariances (M3), and (4) MT model with GE (M4). Several prediction problems were analyzed for each of the GP accuracy of the 4 models. Results of the prediction of traits in HEL, the high yield potential testing site in 2021, show that the best-predicted traits were tuber flesh starch (%), weight of tuber above 60 or below 40 mm in size, and the total tuber weight. In terms of GP, accuracy model M4 gave the best prediction accuracy in 3 traits, namely tuber weight of 40-50 or above 60 mm in size, and total tuber weight, and very similar in the starch trait. For MOS in 2021, the best predictive traits were starch, weight of tubers above 60, 50-60, or below 40 mm in size, and the total tuber weight. MT model M4 was the best GP model based on its accuracy when some cultivars are observed in some traits. For the GP accuracy of traits in UM in 2021, the best predictive traits were the weight of tubers above 60, 50-60, or below 40 mm in size, and the best model was MT M4, followed by models ST M3 and M2.

The Global Assessment of Oilseed Brassica Crop Species Yield, Yield Stability and the Underlying Genetics

The global demand for oilseeds is increasing along with the human population. The family of Brassicaceae crops are no exception, typically harvested as a valuable source of oil, rich in beneficial molecules important for human health. The global capacity for improving Brassica yield has steadily risen over the last 50 years, with the major crop Brassica napus (rapeseed, canola) production increasing to ~72 Gt in 2020. In contrast, the production of Brassica mustard crops has fluctuated, rarely improving in farming efficiency. The drastic increase in global yield of B. napus is largely due to the demand for a stable source of cooking oil. Furthermore, with the adoption of highly efficient farming techniques, yield enhancement programs, breeding programs, the integration of high-throughput phenotyping technology and establishing the underlying genetics, B. napus yields have increased by >450 fold since 1978. Yield stability has been improved with new management strategies targeting diseases and pests, as well as by understanding the complex interaction of environment, phenotype and genotype. This review assesses the global yield and yield stability of agriculturally important oilseed Brassica species and discusses how contemporary farming and genetic techniques have driven improvements.

Crop phenotype prediction using biclustering to explain genotype-by-environment interactions

Phenotypic variation in plants is attributed to genotype (G), environment (E), and genotype-by-environment interaction (GEI). Although the main effects of G and E are typically larger and easier to model, the GEI interaction effects are important and a critical factor when considering such issues as to why some genotypes perform consistently well across a range of environments. In plant breeding, a major challenge is limited information, including a single genotype is tested in only a small subset of all possible test environments. The two-way table of phenotype responses will therefore commonly contain missing data. In this paper, we propose a new model of GEI effects that only requires an input of a two-way table of phenotype observations, with genotypes as rows and environments as columns that do not assume the completeness of data. Our analysis can deal with this scenario as it utilizes a novel biclustering algorithm that can handle missing values, resulting in an output of homogeneous cells with no interactions between G and E. In other words, we identify subsets of genotypes and environments where phenotype can be modeled simply. Based on this, we fit no-interaction models to predict phenotypes of a given crop and draw insights into how a particular cultivar will perform in the unused test environments. Our new methodology is validated on data from different plant species and phenotypes and shows superior performance compared to well-studied statistical approaches.

Association Studies and Genomic Prediction for Genetic Improvements in Agriculture

To feed the fast growing global population with sufficient food using limited global resources, it is urgent to develop and utilize cutting-edge technologies and improve efficiency of agricultural production. In this review, we specifically introduce the concepts, theories, methods, applications and future implications of association studies and predicting unknown genetic value or future phenotypic events using genomics in the area of breeding in agriculture. Genome wide association studies can identify the quantitative genetic loci associated with phenotypes of importance in agriculture, while genomic prediction utilizes individual genetic value to rank selection candidates to improve the next generation of plants or animals. These technologies and methods have improved the efficiency of genetic improvement programs for agricultural production via elite animal breeds and plant varieties. With the development of new data acquisition technologies, there will be more and more data collected from high-through-put technologies to assist agricultural breeding. It will be crucial to extract useful information among these large amounts of data and to face this challenge, more efficient algorithms need to be developed and utilized for analyzing these data. Such development will require knowledge from multiple disciplines of research.

NeuralLasso: Neural Networks Meet Lasso in Genomic Prediction

Prediction of complex traits based on genome-wide marker information is of central importance for both animal and plant breeding. Numerous models have been proposed for the prediction of complex traits and still considerable effort has been given to improve the prediction accuracy of these models, because various genetics factors like additive, dominance and epistasis effects can influence of the prediction accuracy of such models. Recently machine learning (ML) methods have been widely applied for prediction in both animal and plant breeding programs. In this study, we propose a new algorithm for genomic prediction which is based on neural networks, but incorporates classical elements of LASSO. Our new method is able to account for the local epistasis (higher order interaction between the neighboring markers) in the prediction. We compare the prediction accuracy of our new method with the most commonly used prediction methods, such as BayesA, BayesB, Bayesian Lasso (BL), genomic BLUP and Elastic Net (EN) using the heterogenous stock mouse and rice field data sets.

A survey of few-shot learning in smart agriculture: developments, applications, and challenges

With the rise of artificial intelligence, deep learning is gradually applied to the field of agriculture and plant science. However, the excellent performance of deep learning needs to be established on massive numbers of samples. In the field of plant science and biology, it is not easy to obtain a large amount of labeled data. The emergence of few-shot learning solves this problem. It imitates the ability of humans' rapid learning and can learn a new task with only a small number of labeled samples, which greatly reduces the time cost and financial resources. At present, the advanced few-shot learning methods are mainly divided into four categories based on: data augmentation, metric learning, external memory, and parameter optimization, solving the over-fitting problem from different viewpoints. This review comprehensively expounds on few-shot learning in smart agriculture, introduces the definition of few-shot learning, four kinds of learning methods, the publicly available datasets for few-shot learning, various applications in smart agriculture, and the challenges in smart agriculture in future development.

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.

Prospectus of Genomic Selection and Phenomics in Cereal, Legume and Oilseed Breeding Programs

The last decade witnessed an unprecedented increase in the adoption of genomic selection (GS) and phenomics tools in plant breeding programs, especially in major cereal crops. GS has demonstrated the potential for selecting superior genotypes with high precision and accelerating the breeding cycle. Phenomics is a rapidly advancing domain to alleviate phenotyping bottlenecks and explores new large-scale phenotyping and data acquisition methods. In this review, we discuss the lesson learned from GS and phenomics in six self-pollinated crops, primarily focusing on rice, wheat, soybean, common bean, chickpea, and groundnut, and their implementation schemes are discussed after assessing their impact in the breeding programs. Here, the status of the adoption of genomics and phenomics is provided for those crops, with a complete GS overview. GS’s progress until 2020 is discussed in detail, and relevant information and links to the source codes are provided for implementing this technology into plant breeding programs, with most of the examples from wheat breeding programs. Detailed information about various phenotyping tools is provided to strengthen the field of phenomics for a plant breeder in the coming years. Finally, we highlight the benefits of merging genomic selection, phenomics, and machine and deep learning that have resulted in extraordinary results during recent years in wheat, rice, and soybean. Hence, there is a potential for adopting these technologies into crops like the common bean, chickpea, and groundnut. The adoption of phenomics and GS into different breeding programs will accelerate genetic gain that would create an impact on food security, realizing the need to feed an ever-growing population.

Genomic Prediction Methods Accounting for Nonadditive Genetic Effects

The use of genomic information for prediction of future phenotypes or breeding values for the candidates to selection has become a standard over the last decade. However, most procedures for genomic prediction only consider the additive (or substitution) effects associated with polymorphic markers. Nevertheless, the implementation of models that consider nonadditive genetic variation may be interesting because they (1) may increase the ability of prediction, (2) can be used to define mate allocation procedures in plant and animal breeding schemes, and (3) can be used to benefit from nonadditive genetic variation in crossbreeding or purebred breeding schemes. This study reviews the available methods for incorporating nonadditive effects into genomic prediction procedures and their potential applications in predicting future phenotypic performance, mate allocation, and crossbred and purebred selection. Finally, a brief outline of some future research lines is also proposed.

Accounting for Correlation Between Traits in Genomic Prediction

Genomic enabled prediction is playing a key role for the success of genomic selection (GS). However, according to the No Free Lunch Theorem, there is not a universal model that performs well for all data sets. Due to this, many statistical and machine learning models are available for genomic prediction. When multitrait data is available, models that are able to account for correlations between phenotypic traits are preferred, since these models help increase the prediction accuracy when the degree of correlation is moderate to large. For this reason, in this chapter we review multitrait models for genome-enabled prediction and we illustrate the power of this model with real examples. In addition, we provide details of the software (R code) available for its application to help users implement these models with its own data. The multitrait models were implemented under conventional Bayesian Ridge regression and best linear unbiased predictor, but also under a deep learning framework. The multitrait deep learning framework helps implement prediction models with mixed outcomes (continuous, binary, ordinal, and count, measured on different scales), which is not easy in conventional statistical models. The illustrative examples are very detailed in order to make the implementation of multitrait models in plant and animal breeding friendlier for breeders and scientists.

Genome and Environment Based Prediction Models and Methods of Complex Traits Incorporating Genotype × Environment Interaction

Genomic-enabled prediction models are of paramount importance for the successful implementation of genomic selection (GS) based on breeding values. As opposed to animal breeding, plant breeding includes extensive multienvironment and multiyear field trial data. Hence, genomic-enabled prediction models should include genotype × environment (G × E) interaction, which most of the time increases the prediction performance when the response of lines are different from environment to environment. In this chapter, we describe a historical timeline since 2012 related to advances of the GS models that take into account G × E interaction. We describe theoretical and practical aspects of those GS models, including the gains in prediction performance when including G × E structures for both complex continuous and categorical scale traits. Then, we detailed and explained the main G × E genomic prediction models for complex traits measured in continuous and noncontinuous (categorical) scale. Related to G × E interaction models this review also examine the analyses of the information generated with high-throughput phenotype data (phenomic) and the joint analyses of multitrait and multienvironment field trial data that is also employed in the general assessment of multitrait G × E interaction. The inclusion of nongenomic data in increasing the accuracy and biological reliability of the G × E approach is also outlined. We show the recent advances in large-scale envirotyping (enviromics), and how the use of mechanistic computational modeling can derive the crop growth and development aspects useful for predicting phenotypes and explaining G × E.

A New Deep Learning Calibration Method Enhances Genome-Based Prediction of Continuous Crop Traits

Genomic selection (GS) has the potential to revolutionize predictive plant breeding. A reference population is phenotyped and genotyped to train a statistical model that is used to perform genome-enabled predictions of new individuals that were only genotyped. In this vein, deep neural networks, are a type of machine learning model and have been widely adopted for use in GS studies, as they are not parametric methods, making them more adept at capturing nonlinear patterns. However, the training process for deep neural networks is very challenging due to the numerous hyper-parameters that need to be tuned, especially when imperfect tuning can result in biased predictions. In this paper we propose a simple method for calibrating (adjusting) the prediction of continuous response variables resulting from deep learning applications. We evaluated the proposed deep learning calibration method (DL_M2) using four crop breeding data sets and its performance was compared with the standard deep learning method (DL_M1), as well as the standard genomic Best Linear Unbiased Predictor (GBLUP). While the GBLUP was the most accurate model overall, the proposed deep learning calibration method (DL_M2) helped increase the genome-enabled prediction performance in all data sets when compared with the traditional DL method (DL_M1). Taken together, we provide evidence for extending the use of the proposed calibration method to evaluate its potential and consistency for predicting performance in the context of GS applied to plant breeding.

Predicting phenotypes from genetic, environment, management, and historical data using CNNs

Convolutional Neural Networks (CNNs) can perform similarly or better than standard genomic prediction methods when sufficient genetic, environmental, and management data are provided. Predicting phenotypes from genetic (G), environmental (E), and management (M) conditions is a long-standing challenge with implications to agriculture, medicine, and conservation. Most methods reduce the factors in a dataset (feature engineering) in a subjective and potentially oversimplified manner. Deep neural networks such as Multilayer Perceptrons (MPL) and Convolutional Neural Networks (CNN) can overcome this by allowing the data itself to determine which factors are most important. CNN models were developed for predicting agronomic yield from a combination of replicated trials and historical yield survey data. The results were more accurate than standard methods when tested on held-out G, E, and M data (r = 0.50 vs. r = 0.43), and performed slightly worse than standard methods when only G was held out (r = 0.74 vs. r = 0.80). Pre-training on historical data increased accuracy compared to trial data alone. Saliency map analysis indicated the CNN has "learned" to prioritize many factors of known agricultural importance.

Artificial intelligence: A powerful paradigm for scientific research

Artificial intelligence (AI) coupled with promising machine learning (ML) techniques well known from computer science is broadly affecting many aspects of various fields including science and technology, industry, and even our day-to-day life. The ML techniques have been developed to analyze high-throughput data with a view to obtaining useful insights, categorizing, predicting, and making evidence-based decisions in novel ways, which will promote the growth of novel applications and fuel the sustainable booming of AI. This paper undertakes a comprehensive survey on the development and application of AI in different aspects of fundamental sciences, including information science, mathematics, medical science, materials science, geoscience, life science, physics, and chemistry. The challenges that each discipline of science meets, and the potentials of AI techniques to handle these challenges, are discussed in detail. Moreover, we shed light on new research trends entailing the integration of AI into each scientific discipline. The aim of this paper is to provide a broad research guideline on fundamental sciences with potential infusion of AI, to help motivate researchers to deeply understand the state-of-the-art applications of AI-based fundamental sciences, and thereby to help promote the continuous development of these fundamental sciences.

The application of pangenomics and machine learning in genomic selection in plants

Genomic selection approaches have increased the speed of plant breeding, leading to growing crop yields over the last decade. However, climate change is impacting current and future yields, resulting in the need to further accelerate breeding efforts to cope with these changing conditions. Here we present approaches to accelerate plant breeding and incorporate nonadditive effects in genomic selection by applying state-of-the-art machine learning approaches. These approaches are made more powerful by the inclusion of pangenomes, which represent the entire genome content of a species. Understanding the strengths and limitations of machine learning methods, compared with more traditional genomic selection efforts, is paramount to the successful application of these methods in crop breeding. We describe examples of genomic selection and pangenome-based approaches in crop breeding, discuss machine learning-specific challenges, and highlight the potential for the application of machine learning in genomic selection. We believe that careful implementation of machine learning approaches will support crop improvement to help counter the adverse outcomes of climate change on crop production.

Multitrait machine- and deep-learning models for genomic selection using spectral information in a wheat breeding program

Prediction of breeding values is central to plant breeding and has been revolutionized by the adoption of genomic selection (GS). Use of machine- and deep-learning algorithms applied to complex traits in plants can improve prediction accuracies. Because of the tremendous increase in collected data in breeding programs and the slow rate of genetic gain increase, it is required to explore the potential of artificial intelligence in analyzing the data. The main objectives of this study include optimization of multitrait (MT) machine- and deep-learning models for predicting grain yield and grain protein content in wheat (Triticum aestivum L.) using spectral information. This study compares the performance of four machine- and deep-learning-based unitrait (UT) and MT models with traditional genomic best linear unbiased predictor (GBLUP) and Bayesian models. The dataset consisted of 650 recombinant inbred lines (RILs) from a spring wheat breeding program grown for three years (2014-2016), and spectral data were collected at heading and grain filling stages. The MT-GS models performed 0-28.5 and -0.04 to 15% superior to the UT-GS models. Random forest and multilayer perceptron were the best performing machine- and deep-learning models to predict both traits. Four explored Bayesian models gave similar accuracies, which were less than machine- and deep-learning-based models and required increased computational time. Green normalized difference vegetation index (GNDVI) best predicted grain protein content in seven out of the nine MT-GS models. Overall, this study concluded that machine- and deep-learning-based MT-GS models increased prediction accuracy and should be employed in large-scale breeding programs.

Application of a Poisson deep neural network model for the prediction of count data in genome-based prediction

Genomic selection (GS) is revolutionizing conventional ways of developing new plants and animals. However, because it is a predictive methodology, GS strongly depends on statistical and machine learning to perform these predictions. For continuous outcomes, more models are available for GS. Unfortunately, for count data outcomes, there are few efficient statistical machine learning models for large datasets or for datasets with fewer observations than independent variables. For this reason, in this paper, we applied the univariate version of the Poisson deep neural network (PDNN) proposed earlier for genomic predictions of count data. The model was implemented with (a) the negative log-likelihood of Poisson distribution as the loss function, (b) the rectified linear activation unit as the activation function in hidden layers, and (c) the exponential activation function in the output layer. The advantage of the PDNN model is that it captures complex patterns in the data by implementing many nonlinear transformations in the hidden layers. Moreover, since it was implemented in Tensorflow as the back-end, and in Keras as the front-end, the model can be applied to moderate and large datasets, which is a significant advantage over previous GS models for count data. The PDNN model was compared with deep learning models with continuous outcomes, conventional generalized Poisson regression models, and conventional Bayesian regression methods. We found that the PDNN model outperformed the Bayesian regression and generalized Poisson regression methods in terms of prediction accuracy, although it was not better than the conventional deep neural network with continuous outcomes.

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.

Next-Generation Breeding Strategies for Climate-Ready Crops

Climate change is a threat to global food security due to the reduction of crop productivity around the globe. Food security is a matter of concern for stakeholders and policymakers as the global population is predicted to bypass 10 billion in the coming years. Crop improvement via modern breeding techniques along with efficient agronomic practices innovations in microbiome applications, and exploiting the natural variations in underutilized crops is an excellent way forward to fulfill future food requirements. In this review, we describe the next-generation breeding tools that can be used to increase crop production by developing climate-resilient superior genotypes to cope with the future challenges of global food security. Recent innovations in genomic-assisted breeding (GAB) strategies allow the construction of highly annotated crop pan-genomes to give a snapshot of the full landscape of genetic diversity (GD) and recapture the lost gene repertoire of a species. Pan-genomes provide new platforms to exploit these unique genes or genetic variation for optimizing breeding programs. The advent of next-generation clustered regularly interspaced short palindromic repeat/CRISPR-associated (CRISPR/Cas) systems, such as prime editing, base editing, and de nova domestication, has institutionalized the idea that genome editing is revamped for crop improvement. Also, the availability of versatile Cas orthologs, including Cas9, Cas12, Cas13, and Cas14, improved the editing efficiency. Now, the CRISPR/Cas systems have numerous applications in crop research and successfully edit the major crop to develop resistance against abiotic and biotic stress. By adopting high-throughput phenotyping approaches and big data analytics tools like artificial intelligence (AI) and machine learning (ML), agriculture is heading toward automation or digitalization. The integration of speed breeding with genomic and phenomic tools can allow rapid gene identifications and ultimately accelerate crop improvement programs. In addition, the integration of next-generation multidisciplinary breeding platforms can open exciting avenues to develop climate-ready crops toward global food security.

Genomic Selection for End-Use Quality and Processing Traits in Soft White Winter Wheat Breeding Program with Machine and Deep Learning Models

Breeding for grain yield, biotic and abiotic stress resistance, and end-use quality are important goals of wheat breeding programs. Screening for end-use quality traits is usually secondary to grain yield due to high labor needs, cost of testing, and large seed requirements for phenotyping. Genomic selection provides an alternative to predict performance using genome-wide markers under forward and across location predictions, where a previous year's dataset can be used to build the models. Due to large datasets in breeding programs, we explored the potential of the machine and deep learning models to predict fourteen end-use quality traits in a winter wheat breeding program. The population used consisted of 666 wheat genotypes screened for five years (2015-19) at two locations (Pullman and Lind, WA, USA). Nine different models, including two machine learning (random forest and support vector machine) and two deep learning models (convolutional neural network and multilayer perceptron) were explored for cross-validation, forward, and across locations predictions. The prediction accuracies for different traits varied from 0.45-0.81, 0.29-0.55, and 0.27-0.50 under cross-validation, forward, and across location predictions. In general, forward prediction accuracies kept increasing over time due to increments in training data size and was more evident for machine and deep learning models. Deep learning models were superior over the traditional ridge regression best linear unbiased prediction (RRBLUP) and Bayesian models under all prediction scenarios. The high accuracy observed for end-use quality traits in this study support predicting them in early generations, leading to the advancement of superior genotypes to more extensive grain yield trails. Furthermore, the superior performance of machine and deep learning models strengthens the idea to include them in large scale breeding programs for predicting complex traits.

Harnessing translational research in wheat for climate resilience

Despite being the world's most widely grown crop, research investments in wheat (Triticum aestivum and Triticum durum) fall behind those in other staple crops. Current yield gains will not meet 2050 needs, and climate stresses compound this challenge. However, there is good evidence that heat and drought resilience can be boosted through translating promising ideas into novel breeding technologies using powerful new tools in genetics and remote sensing, for example. Such technologies can also be applied to identify climate resilience traits from among the vast and largely untapped reserve of wheat genetic resources in collections worldwide. This review describes multi-pronged research opportunities at the focus of the Heat and Drought Wheat Improvement Consortium (coordinated by CIMMYT), which together create a pipeline to boost heat and drought resilience, specifically: improving crop design targets using big data approaches; developing phenomic tools for field-based screening and research; applying genomic technologies to elucidate the bases of climate resilience traits; and applying these outputs in developing next-generation breeding methods. The global impact of these outputs will be validated through the International Wheat Improvement Network, a global germplasm development and testing system that contributes key productivity traits to approximately half of the global wheat-growing area.

Machine learning in plant science and plant breeding

Technological developments have revolutionized measurements on plant genotypes and phenotypes, leading to routine production of large, complex data sets. This has led to increased efforts to extract meaning from these measurements and to integrate various data sets. Concurrently, machine learning has rapidly evolved and is now widely applied in science in general and in plant genotyping and phenotyping in particular. Here, we review the application of machine learning in the context of plant science and plant breeding. We focus on analyses at different phenotype levels, from biochemical to yield, and in connecting genotypes to these. In this way, we illustrate how machine learning offers a suite of methods that enable researchers to find meaningful patterns in relevant plant data.

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.

Genome-Wide Prediction of Complex Traits in Two Outcrossing Plant Species Through Deep Learning and Bayesian Regularized Neural Network

Genomic selection models were investigated to predict several complex traits in breeding populations of Zea mays L. and Eucalyptus globulus Labill. For this, the following methods of Machine Learning (ML) were implemented: (i) Deep Learning (DL) and (ii) Bayesian Regularized Neural Network (BRNN) both in combination with different hyperparameters. These ML methods were also compared with Genomic Best Linear Unbiased Prediction (GBLUP) and different Bayesian regression models [Bayes A, Bayes B, Bayes Cπ, Bayesian Ridge Regression, Bayesian LASSO, and Reproducing Kernel Hilbert Space (RKHS)]. DL models, using Rectified Linear Units (as the activation function), had higher predictive ability values, which varied from 0.27 (pilodyn penetration of 6 years old eucalypt trees) to 0.78 (flowering-related traits of maize). Moreover, the larger mini-batch size (100%) had a significantly higher predictive ability for wood-related traits than the smaller mini-batch size (10%). On the other hand, in the BRNN method, the architectures of one and two layers that used only the pureline function showed better results of prediction, with values ranging from 0.21 (pilodyn penetration) to 0.71 (flowering traits). A significant increase in the prediction ability was observed for DL in comparison with other methods of genomic prediction (Bayesian alphabet models, GBLUP, RKHS, and BRNN). Another important finding was the usefulness of DL models (through an iterative algorithm) as an SNP detection strategy for genome-wide association studies. The results of this study confirm the importance of DL for genome-wide analyses and crop/tree improvement strategies, which holds promise for accelerating breeding progress.

Using Local Convolutional Neural Networks for Genomic Prediction

The prediction of breeding values and phenotypes is of central importance for both livestock and crop breeding. In this study, we analyze the use of artificial neural networks (ANN) and, in particular, local convolutional neural networks (LCNN) for genomic prediction, as a region-specific filter corresponds much better with our prior genetic knowledge on the genetic architecture of traits than traditional convolutional neural networks. Model performances are evaluated on a simulated maize data panel (n = 10,000; p = 34,595) and real Arabidopsis data (n = 2,039; p = 180,000) for a variety of traits based on their predictive ability. The baseline LCNN, containing one local convolutional layer (kernel size: 10) and two fully connected layers with 64 nodes each, is outperforming commonly proposed ANNs (multi layer perceptrons and convolutional neural networks) for basically all considered traits. For traits with high heritability and large training population as present in the simulated data, LCNN are even outperforming state-of-the-art methods like genomic best linear unbiased prediction (GBLUP), Bayesian models and extended GBLUP, indicated by an increase in predictive ability of up to 24%. However, for small training populations, these state-of-the-art methods outperform all considered ANNs. Nevertheless, the LCNN still outperforms all other considered ANNs by around 10%. Minor improvements to the tested baseline network architecture of the LCNN were obtained by increasing the kernel size and of reducing the stride, whereas the number of subsequent fully connected layers and their node sizes had neglectable impact. Although gains in predictive ability were obtained for large scale data sets by using LCNNs, the practical use of ANNs comes with additional problems, such as the need of genotyping all considered individuals, the lack of estimation of heritability and reliability. Furthermore, breeding values are additive by design, whereas ANN-based estimates are not. However, ANNs also comes with new opportunities, as networks can easily be extended to account for additional inputs (omics, weather etc.) and outputs (multi-trait models), and computing time increases linearly with the number of individuals. With advances in high-throughput phenotyping and cheaper genotyping, ANNs can become a valid alternative for genomic prediction.

A Multivariate Poisson Deep Learning Model for Genomic Prediction of Count Data

The paradigm called genomic selection (GS) is a revolutionary way of developing new plants and animals. This is a predictive methodology, since it uses learning methods to perform its task. Unfortunately, there is no universal model that can be used for all types of predictions; for this reason, specific methodologies are required for each type of output (response variables). Since there is a lack of efficient methodologies for multivariate count data outcomes, in this paper, a multivariate Poisson deep neural network (MPDN) model is proposed for the genomic prediction of various count outcomes simultaneously. The MPDN model uses the minus log-likelihood of a Poisson distribution as a loss function, in hidden layers for capturing nonlinear patterns using the rectified linear unit (RELU) activation function and, in the output layer, the exponential activation function was used for producing outputs on the same scale of counts. The proposed MPDN model was compared to conventional generalized Poisson regression models and univariate Poisson deep learning models in two experimental data sets of count data. We found that the proposed MPDL outperformed univariate Poisson deep neural network models, but did not outperform, in terms of prediction, the univariate generalized Poisson regression models. All deep learning models were implemented in Tensorflow as back-end and Keras as front-end, which allows implementing these models on moderate and large data sets, which is a significant advantage over previous GS models for multivariate count data.

Genome-based prediction of multiple wheat quality traits in multiple years

Wheat quality improvement is an important objective in all wheat breeding programs. However, due to the cost, time and quantity of seed required, wheat quality is typically analyzed only in the last stages of the breeding cycle on a limited number of samples. The use of genomic prediction could greatly help to select for wheat quality more efficiently by reducing the cost and time required for this analysis. Here were evaluated the prediction performances of 13 wheat quality traits under two multi-trait models (Bayesian multi-trait multi-environment [BMTME] and multi-trait ridge regression [MTR]) using five data sets of wheat lines evaluated in the field during two consecutive years. Lines in the second year (testing) were predicted using the quality information obtained in the first year (training). For most quality traits were found moderate to high prediction accuracies, suggesting that the use of genomic selection could be feasible. The best predictions were obtained with the BMTME model in all traits and the worst with the MTR model. The best predictions with the BMTME model under the mean arctangent absolute percentage error (MAAPE) were for test weight across the five data sets, whereas the worst predictions were for the alveograph trait ALVPL. In contrast, under Pearson's correlation, the best predictions depended on the data set. The results obtained suggest that the BMTME model should be preferred for multi-trait prediction analyses. This model allows to obtain not only the correlation among traits, but also the correlation among environments, helping to increase the prediction accuracy.

Application of Genomic Big Data in Plant Breeding:Past, Present, and Future

Plant breeding has a long history of developing new varieties that have ensured the food security of the human population. During this long journey together with humanity, plant breeders have successfully integrated the latest innovations in science and technologies to accelerate the increase in crop production and quality. For the past two decades, since the completion of human genome sequencing, genomic tools and sequencing technologies have advanced remarkably, and adopting these innovations has enabled us to cost down and/or speed up the plant breeding process. Currently, with the growing mass of genomic data and digitalized biological data, interdisciplinary approaches using new technologies could lead to a new paradigm of plant breeding. In this review, we summarize the overall history and advances of plant breeding, which have been aided by plant genomic research. We highlight the key advances in the field of plant genomics that have impacted plant breeding over the past decades and introduce the current status of innovative approaches such as genomic selection, which could overcome limitations of conventional breeding and enhance the rate of genetic gain.

Genome-based prediction of Bayesian linear and non-linear regression models for ordinal data

Linear and non-linear models used in applications of genomic selection (GS) can fit different types of responses (e.g., continuous, ordinal, binary). In recent years, several genomic-enabled prediction models have been developed for predicting complex traits in genomic-assisted animal and plant breeding. These models include linear, non-linear and non-parametric models, mostly for continuous responses and less frequently for categorical responses. Several linear and non-linear models are special cases of a more general family of statistical models known as artificial neural networks, which provide better prediction ability than other models. In this paper, we propose a Bayesian Regularized Neural Network (BRNNO) for modelling ordinal data. The proposed model was fitted using a Bayesian framework; we used the data augmentation algorithm to facilitate computations. The proposed model was fitted using the Gibbs Maximum a Posteriori and Generalized EM algorithm implemented by combining code written in C and R programming languages. The new model was tested with two real maize datasets evaluated for Septoria and GLS diseases and was compared with the Bayesian Ordered Probit Model (BOPM). Results indicated that the BRNNO model performed better in terms of genomic-based prediction than the BOPM model.

Deep learning models in genomics; are we there yet?

With the evolution of biotechnology and the introduction of the high throughput sequencing, researchers have the ability to produce and analyze vast amounts of genomics data. Since genomics produce big data, most of the bioinformatics algorithms are based on machine learning methodologies, and lately deep learning, to identify patterns, make predictions and model the progression or treatment of a disease. Advances in deep learning created an unprecedented momentum in biomedical informatics and have given rise to new bioinformatics and computational biology research areas. It is evident that deep learning models can provide higher accuracies in specific tasks of genomics than the state of the art methodologies. Given the growing trend on the application of deep learning architectures in genomics research, in this mini review we outline the most prominent models, we highlight possible pitfalls and discuss future directions. We foresee deep learning accelerating changes in the area of genomics, especially for multi-scale and multimodal data analysis for precision medicine.

Integrating High-Throughput Phenotyping and Statistical Genomic Methods to Genetically Improve Longitudinal Traits in Crops

The rapid development of remote sensing in agronomic research allows the dynamic nature of longitudinal traits to be adequately described, which may enhance the genetic improvement of crop efficiency. For traits such as light interception, biomass accumulation, and responses to stressors, the data generated by the various high-throughput phenotyping (HTP) methods requires adequate statistical techniques to evaluate phenotypic records throughout time. As a consequence, information about plant functioning and activation of genes, as well as the interaction of gene networks at different stages of plant development and in response to environmental stimulus can be exploited. In this review, we outline the current analytical approaches in quantitative genetics that are applied to longitudinal traits in crops throughout development, describe the advantages and pitfalls of each approach, and indicate future research directions and opportunities.

Emerging Advanced Technologies to Mitigate the Impact of Climate Change in Africa

Agriculture remains critical to Africa's socioeconomic development, employing 65% of the work force and contributing 32% of GDP (Gross Domestic Product). Low productivity, which characterises food production in many Africa countries, remains a major concern. Compounded by the effects of climate change and lack of technical expertise, recent reports suggest that the impacts of climate change on agriculture and food systems in African countries may have further-reaching consequences than previously anticipated. Thus, it has become imperative that African scientists and farmers adopt new technologies which facilitate their research and provide smart agricultural solutions to mitigating current and future climate change-related challenges. Advanced technologies have been developed across the globe to facilitate adaptation to climate change in the agriculture sector. Clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR-associated protein 9 (Cas9), synthetic biology, and genomic selection, among others, constitute examples of some of these technologies. In this work, emerging advanced technologies with the potential to effectively mitigate climate change in Africa are reviewed. The authors show how these technologies can be utilised to enhance knowledge discovery for increased production in a climate change-impacted environment. We conclude that the application of these technologies could empower African scientists to explore agricultural strategies more resilient to the effects of climate change. Additionally, we conclude that support for African scientists from the international community in various forms is necessary to help Africans avoid the full undesirable effects of climate change.

Exploring Deep Learning for Complex Trait Genomic Prediction in Polyploid Outcrossing Species

Genomic prediction (GP) is the procedure whereby the genetic merits of untested candidates are predicted using genome wide marker information. Although numerous examples of GP exist in plants and animals, applications to polyploid organisms are still scarce, partly due to limited genome resources and the complexity of this system. Deep learning (DL) techniques comprise a heterogeneous collection of machine learning algorithms that have excelled at many prediction tasks. A potential advantage of DL for GP over standard linear model methods is that DL can potentially take into account all genetic interactions, including dominance and epistasis, which are expected to be of special relevance in most polyploids. In this study, we evaluated the predictive accuracy of linear and DL techniques in two important small fruits or berries: strawberry and blueberry. The two datasets contained a total of 1,358 allopolyploid strawberry (2n=8x=112) and 1,802 autopolyploid blueberry (2n=4x=48) individuals, genotyped for 9,908 and 73,045 single nucleotide polymorphism (SNP) markers, respectively, and phenotyped for five agronomic traits each. DL depends on numerous parameters that influence performance and optimizing hyperparameter values can be a critical step. Here we show that interactions between hyperparameter combinations should be expected and that the number of convolutional filters and regularization in the first layers can have an important effect on model performance. In terms of genomic prediction, we did not find an advantage of DL over linear model methods, except when the epistasis component was important. Linear Bayesian models were better than convolutional neural networks for the full additive architecture, whereas the opposite was observed under strong epistasis. However, by using a parameterization capable of taking into account these non-linear effects, Bayesian linear models can match or exceed the predictive accuracy of DL. A semiautomatic implementation of the DL pipeline is available at https://github.com/lauzingaretti/deepGP/.

A Multiple-Trait Bayesian Lasso for Genome-Enabled Analysis and Prediction of Complex Traits

A multiple-trait Bayesian LASSO (MBL) for genome-based analysis and prediction of quantitative traits is presented and applied to two real data sets. The data-generating model is a multivariate linear Bayesian regression on possibly a huge number of molecular markers, and with a Gaussian residual distribution posed. Each (one per marker) of the [Formula: see text] vectors of regression coefficients (T: number of traits) is assigned the same T-variate Laplace prior distribution, with a null mean vector and unknown scale matrix Σ. The multivariate prior reduces to that of the standard univariate Bayesian LASSO when [Formula: see text] The covariance matrix of the residual distribution is assigned a multivariate Jeffreys prior, and Σ is given an inverse-Wishart prior. The unknown quantities in the model are learned using a Markov chain Monte Carlo sampling scheme constructed using a scale-mixture of normal distributions representation. MBL is demonstrated in a bivariate context employing two publicly available data sets using a bivariate genomic best linear unbiased prediction model (GBLUP) for benchmarking results. The first data set is one where wheat grain yields in two different environments are treated as distinct traits. The second data set comes from genotyped Pinus trees, with each individual measured for two traits: rust bin and gall volume. In MBL, the bivariate marker effects are shrunk differentially, i.e., "short" vectors are more strongly shrunk toward the origin than in GBLUP; conversely, "long" vectors are shrunk less. A predictive comparison was carried out as well in wheat, where the comparators of MBL were bivariate GBLUP and bivariate Bayes Cπ-a variable selection procedure. A training-testing layout was used, with 100 random reconstructions of training and testing sets. For the wheat data, all methods produced similar predictions. In Pinus, MBL gave better predictions that either a Bayesian bivariate GBLUP or the single trait Bayesian LASSO. MBL has been implemented in the Julia language package JWAS, and is now available for the scientific community to explore with different traits, species, and environments. It is well known that there is no universally best prediction machine, and MBL represents a new resource in the armamentarium for genome-enabled analysis and prediction of complex traits.

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.

Multi-Trait, Multi-Environment Genomic Prediction of Durum Wheat With Genomic Best Linear Unbiased Predictor and Deep Learning Methods

Although durum wheat (Triticum turgidum var. durum Desf.) is a minor cereal crop representing just 5-7% of the world's total wheat crop, it is a staple food in Mediterranean countries, where it is used to produce pasta, couscous, bulgur and bread. In this paper, we cover multi-trait prediction of grain yield (GY), days to heading (DH) and plant height (PH) of 270 durum wheat lines that were evaluated in 43 environments (country-location-year combinations) across a broad range of water regimes in the Mediterranean Basin and other locations. Multi-trait prediction analyses were performed by implementing a multi-trait deep learning model (MTDL) with a feed-forward network topology and a rectified linear unit activation function with a grid search approach for the selection of hyper-parameters. The results of the multi-trait deep learning method were also compared with univariate predictions of the genomic best linear unbiased predictor (GBLUP) method and the univariate counterpart of the multi-trait deep learning method (UDL). All models were implemented with and without the genotype × environment interaction term. We found that the best predictions were observed without the genotype × environment interaction term in the UDL and MTDL methods. However, under the GBLUP method, the best predictions were observed when the genotype × environment interaction term was taken into account. We also found that in general the best predictions were observed under the GBLUP model; however, the predictions of the MTDL were very similar to those of the GBLUP model. This result provides more evidence that the GBLUP model is a powerful approach for genomic prediction, but also that the deep learning method is a practical approach for predicting univariate and multivariate traits in the context of genomic selection.

Deep Kernel for Genomic and Near Infrared Predictions in Multi-environment Breeding Trials

Kernel methods are flexible and easy to interpret and have been successfully used in genomic-enabled prediction of various plant species. Kernel methods used in genomic prediction comprise the linear genomic best linear unbiased predictor (GBLUP or GB) kernel, and the Gaussian kernel (GK). In general, these kernels have been used with two statistical models: single-environment and genomic × environment (GE) models. Recently near infrared spectroscopy (NIR) has been used as an inexpensive and non-destructive high-throughput phenotyping method for predicting unobserved line performance in plant breeding trials. In this study, we used a non-linear arc-cosine kernel (AK) that emulates deep learning artificial neural networks. We compared AK prediction accuracy with the prediction accuracy of GB and GK kernel methods in four genomic data sets, one of which also includes pedigree and NIR information. Results show that for all four data sets, AK and GK kernels achieved higher prediction accuracy than the linear GB kernel for the single-environment and GE multi-environment models. In addition, AK achieved similar or slightly higher prediction accuracy than the GK kernel. For all data sets, the GE model achieved higher prediction accuracy than the single-environment model. For the data set that includes pedigree, markers and NIR, results show that the NIR wavelength alone achieved lower prediction accuracy than the genomic information alone; however, the pedigree plus NIR information achieved only slightly lower prediction accuracy than the marker plus the NIR high-throughput data.