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

Abstract

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.

Introduction

Enhancing agricultural production efficiency by reducing yield gaps while also breeding more stress-resilient cultivars is the next challenge for plant breeders (Godfray et al., 2010; Foley et al., 2011; Ray et al., 2013; Challinor et al., 2014; Tai et al., 2014). The most feasible solutions are developing innovative approaches to speed up the genetic improvement of economically important traits and characterizing novel traits and incorporating them into breeding programs (Duvick, 2005; Lange and Federizzi, 2009; Fischer and Edmeades, 2010; Nolan and Santos, 2012; Rogers et al., 2015).

Plant breeding was established as a science in the beginning of the 20th century, when new insights about the genetic basis of phenotypic variation for quantitative traits were integrated with the foundational theories of inheritance mechanisms and crop hybridization elucidated by Mendel and Darwin, respectively (Johannsen, 1909, 1911; East, 1911; Bradshaw, 2017). Since then, plant breeders have improved crop productivity by selecting for numerous traits. Breeding objectives are constantly refined to address new challenges, including adaptation to new production areas, addressing emerging pests and diseases, various end uses, advanced farming technologies, and climate change (Toenniessen, 2002; Baenziger et al., 2006; Tester and Langridge, 2010; Gilliham et al., 2017). For instance, in 1955 the focus of soybean breeding was to increase seed oil content, canopy ground cover, and ripening uniformity. Ten years later the focus shifted to reducing pod shattering and lodging, and then it changed again over the years to include quality and value-added traits (Baenziger et al., 2006). Similar shifts in breeding goals and phenotyping technologies have also occurred in animal breeding (Henryon et al., 2014; Miglior et al., 2017). However, throughout history, improving traits of interest depends on the ability to quantify phenotypes across genotypes replicated over multiple environments (Stoskopf et al., 1994). Therefore, potentially valuable traits may have been neglected due to costly phenotyping and technological limitations.

Plant phenotyping has always been paramount for genetic improvement. Recent advances in proximal remote sensing, paired with new sensors and computer science applications, has enabled cost-effective high-throughput phenotyping (HTP) and dissection of novel traits (Montes et al., 2007; Furbank and Tester, 2011; Fiorani and Schurr, 2013; Araus and Cairns, 2014; Coppens et al., 2017). HTP provides time-series measurements that track the development of a crop through its life stages and as it responds to the environment. Information on gene function, the activation of genes, interaction of genes networks at different stages of plant development and in response to environmental stimulus can now be exploited (Wu and Lin, 2006; Montes et al., 2007). It is increasingly possible for plant breeders to consider light interception, biomass accumulation, and response to drought stress as dynamic traits, rather than static points in time (Montes et al., 2007). This analytical framework enhances our understanding of crop development and bridges gaps in understanding the relationship between genotype and phenotype (Granier and Vile, 2014; Araus et al., 2018).

Traits that are expressed repeatedly or continuously over the lifetime of an individual can be defined as longitudinal traits (Yang et al., 2006; Oliveira et al., 2019a), infinite-dimensional traits (Kirkpatrick and Heckman, 1989), or function-valued traits (Promislow et al., 1996). The study of longitudinal traits can provide important insights into the genetic mechanisms underlying physiological responses to environmental stresses and developmental processes. This information can be used to improve predictive ability for complex polygenic traits under multivariate settings, and contribute to identifying overall (e.g., soybean yield) or time specific Quantitative Trait Loci (QTL) (Fahlgren et al., 2015; Campbell et al., 2017; Sun et al., 2017). Such analysis enables assessment of the statistical association of genetic and environmental factors, such as the relationship between molecular markers and response to abiotic stress at different developmental stages (Langridge and Fleury, 2011). In this context, phenotypic data describes a function changing continuously in response to other variables (Stinchcombe and Kirkpatrick, 2012; Granier and Vile, 2014). These approaches generate a vast amount of data, which requires advanced statistical approaches to enable evaluation of the phenotypic data as a function of time. In this literature review, we will outline the current analytical approaches in quantitative genetics and genomics that can be applied to HTP quantified over time (Figure 1). In addition, we describe the advantages and pitfalls of each method and explore directions and opportunities for future research.

FIGURE 1

Phenotyping Longitudinal Traits

Current HTP platforms, also referred to as “phenomic” tools, include a variety of methodologies that use remote sensing to obtain non-destructive phenotypic measurements, either in controlled environments or in the field (Pauli et al., 2016b). The most common types of sensors for crop phenotyping include red-green-blue (RGB; Xavier et al., 2017), multispectral (Xu et al., 2019), hyperspectral (Bodner et al., 2018), fluorescence (Pérez-Bueno et al., 2016), thermal (Sagan et al., 2019), three-dimensional (3D; Topp et al., 2013), and laser-imaging detection and ranging (LiDAR) (Sun et al., 2018) devices. In general, these sensors rely on the interaction between electromagnetic radiation and plants (reflecting, absorbing, or transmitting photons), which is captured by the sensor as reflected radiation (Fiorani et al., 2012; Li et al., 2014). Thus, the sensors interpret the plants as optical objects, with each component of the plant displaying a characteristic spectral signature arising from wavelength-specific properties of absorbance, reflectance, and transmittance from the vegetation surface (Schowengerdt, 2012; Li et al., 2014).

The spectral signatures of plants change during the life cycle, giving rise to genotype-time-specific phenotypes. For instance, during senescence, there is an increase of reflectance in the red region caused by a loss of chlorophyll (Schowengerdt, 2012). For field-based phenotyping, these sensing tools are usually integrated into ground or aerial vehicles (Araus et al., 2018). Most HTP platforms have the spatial and temporal resolution needed to capture longitudinal traits. However, the needs and resources of the specific experiment should drive the choice of platform and sensor, as these choices directly impact the scale and type of research (Pauli et al., 2016b). Several reviews have focused on HTP and its nuances, such as data collection, data processing and types of sensors (Fahlgren et al., 2015; Rahaman et al., 2015; Singh et al., 2016; Tardieu et al., 2017; Yang et al., 2017; Zhao et al., 2019; Reynolds et al., 2020).

The genetic control of longitudinal traits captured with HTP was recently reported for different crops. In a greenhouse, Neilson et al. (2015) investigated the growth and dynamic phenotypic responses of sorghum to water-limited conditions and various levels of fertilizer over time. They defined and measured several traits, including leaf area, shoot biomass, height, tiller number, and leaf greenness using laser scanning, RGB, and near-infrared (NIR) cameras. In barley, multiple sensors captured daily images in a greenhouse over 58 days measuring the spectrum of visible light, fluorescence, and NIR in order to dissect phenotypic components of drought responses (Chen et al., 2014). Tiller growth in rice was examined using more than 700 traits extracted from an imaging system combining computed tomography (CT) and RGB imaging during the tillering process (Wu et al., 2019).

In the field, Sun et al. (2018) used LiDAR mounted on a tractor to quantify leaf area, stem height, and plant volume in cotton in the field from 43 to 109 days after planting. From these measurements, they generated genotype-specific growth curves and assessed variability in traits and their genetic correlation to yield over time. In soybean, an unmanned aerial vehicle (UAV, aka drone) with an RGB camera measured canopy coverage in the field across several days after planting for genome-wide association analysis and genomic selection (Xavier et al., 2017; Hearst, 2019; Moreira et al., 2019). Blancon et al. (2019) characterized the dynamics of the green-leaf-area index in a diversity panel of maize under well-watered and water-deficient treatments using multispectral imagery acquired from a UAV throughout the growth cycle. Thermal and hyperspectral sensors mounted to a manned aircraft were used to extract canopy temperature and vegetation indexes of more than 500 lines of wheat in five field environments over a range of dates (Rutkoski et al., 2016).

Field-based HTP for roots has seen less progress than HTP of above-ground traits due to the difficulty of below-ground imaging (Atkinson et al., 2019). Shovelomics is a high-throughput method for root phenotyping and that has been used for crops, such as maize, common bean, cowpea, and wheat (Trachsel et al., 2011; Burridge et al., 2016; York et al., 2018). It consists of extracting several traits from images. However, it is destructive and labor-intensive, requiring manual root excavation, which limits its ability to capture repeated records over time. Recently, geophysical techniques, including electrical resistance tomography, electromagnetic inductance, and ground-penetrating radar have contributed to identify and quantify roots in the field in a non-destructive manner (Shanahan et al., 2015; Whalley et al., 2017; Liu et al., 2018a; Atkinson et al., 2019). Nevertheless, HTP root phenotyping is more commonly performed in controlled environments that allow the use of alternative growth systems that enable root imaging, such as rhizotrons, growth pouches and transparent artificial growth media (Atkinson et al., 2019; Ma et al., 2019). To model the growth dynamics of maize, Hund et al. (2009) performed daily scans of root systems grown over blotting paper. Topp et al. (2013) used 3D imaging phenotyping of rice root architecture in a gellan gum medium on various days of growth to perform QTL detection analysis. Recent advances in the use of X-ray CT and magnetic resonance imaging in plant sciences have enabled monitoring of root system architecture and dynamic growth over time in soil (Metzner et al., 2015; Pfeifer et al., 2015; van Dusschoten et al., 2016; Pflugfelder et al., 2017; Gao et al., 2019). More details and additional techniques for root phenotyping are presented in Atkinson et al. (2019).

Modeling Longitudinal Traits

Plant growth and development are characterized by several phenotypic changes, which can only be studied by monitoring repeated phenotypes over time (Li and Sillanpää, 2015). HTP platforms allow tracking of traits with a high temporal resolution, whether continuously over time or in discrete intervals (Furbank and Tester, 2011). Traditionally, mathematical functions are used to describe temporal trajectories of traits during the plant’s life cycle (Paine et al., 2012). Analysis of longitudinal traits usually employs one of the following two techniques (Li and Sillanpää, 2013): (1) smooth functions (such as splines; e.g., van Eeuwijk et al., 2018; Oliveira et al., 2019a) or parametric functions (such as growth models; e.g., Paine et al., 2012) to fit the phenotypic records over time, providing interpolated values for all time points; or (2) the data is reparameterized by estimating the function’s coefficients, which are then used in the analysis to represent the trait over time. Either way, it is necessary to select the function that best fits the shape of the trajectory of the trait to accurately estimate the curve parameters and results. Paine et al. (2012) provide a detailed review about growth models, highlighting basic functional forms, advantages and disadvantages. In this section, we will describe the main functions that have been successfully used to fit a variety of traits for the purposes of crop improvement.

Many of the complexities of plant growth are commonly represented using non-linear growth models that account for temporal variation in growth, capturing age and size-dependent growth (Paine et al., 2012). Usually, the growth pattern within a plant life cycle follows a sigmoid curve (S-shaped) characterized by an initial slow growth that then increases rapidly, approaching an exponential growth rate, and finally slows when it reaches a saturation phase (Yin et al., 2003). The S-shaped curve can be described by sigmoidal functions such as the logistic, Gompertz, Richards or β functions (Gompertz, 1815; Richards, 1959; Yin et al., 2003; Poorter et al., 2013). In this case, the Gompertz function is a special case of the Richard function; which is one of the oldest growth models frequently used to fit various biological processes across species (Tjørve and Tjørve, 2017). The Gompertz function has been used to describe biomass accumulation in maize kernels (Meade et al., 2013), barley biomass (Chen et al., 2014), and various longitudinal traits in wheat (Camargo et al., 2018) and sorghum (Neilson et al., 2015). The Logistic function is more commonly used in its asymptotic form to describe the time dependence of biological growth processes for traits such as biomass, canopy coverage, canopy size, volume, length, and area (Thornley et al., 2005; Paine et al., 2012). The Logistic function can have one, two, three, four, or five-parameters (Tessmer et al., 2013). One- and two-parameter logistic models are simplistic and frequently do not fit the data well, but are still used in several studies (e.g., Paine et al., 2012; Tessmer et al., 2013). The three-parameter logistic function (3PL; also known as a Verhulst or autocatalytic growth function) is perhaps the most popular model for plant growth analysis. In a water-limitation experiment in sorghum, a 3PL model had the best performance of a variety of sigmoidal models to fit the projected leaf area (Neilson et al., 2015). Sun et al. (2018) used a 3PL model to fit growth curves for canopy height, projected canopy area, and plant volume obtained from HTP in cotton. In wheat, Baillot et al. (2018) estimated various grain-filling parameters by fitting a 3PL model. The four-parameter logistic (4PL) model is more flexible than the 3PL as it has fewer constraints (Pinheiro and Bates, 2000). Camargo et al. (2018) phenotyped the average of area, height, and senescence in wheat throughout its lifecycle and found that 4PL models best fit the longitudinal data. The five-parameter version (5PL) provides maximum flexibility and accommodates asymmetry (Gottschalk and Dunn, 2005), despite its higher complexity compared to the lower number of parameters.

Many biological curves cannot be described by sigmoidal functions. A Power Law (also known as allometry) function is a type of non-asymptotic, non-linear growth model that does not produce an S-shaped curve (Marquet et al., 2005). They are often used in ecology to predict relationships in plant communities (Chen and Shiyomi, 2019). It effectively captures temporal variation in growth as it allows the relative growth rate to slow down over time and with an increase in biomass (Paine et al., 2012). A Power Law function was used to fit projected leaf area data in sorghum receiving various levels of nitrogen (Neilson et al., 2015), as well as leaf length and rosette area in Arabidopsis thaliana (An et al., 2016).

Linear models, such as orthogonal polynomials and spline functions, are also used to fit longitudinal traits (Oliveira et al., 2019a). The use of polynomials in crop growth models started in the 1960s as a functional approach to fit growth data and provide a clear picture of ontogenetic drift (Vernon and Allison, 1963; Hughes and Freeman, 1967; Poorter, 1989). One of the advantages of these functions is that they do not require prior knowledge of the longitudinal shape of the phenotype. Therefore, they are useful for fitting biological data of any shape simply by choosing different orders of the polynomials. Although they are not linear in time, polynomial functions are linear in their parameters, and consequently, can take advantage of the inference methods available for linear models (Yang et al., 2006). For instance, cubic polynomial functions have been used to describe grain growth in crops such as rice (Jones et al., 1979; Shi et al., 2015), wheat (Gebeyehou et al., 1982), barley (Leon and Geisler, 1994), and safflower (Koutroubas and Papakosta, 2010). One of the main difficulties with this approach is choosing the appropriate degree of polynomial to fit the data while avoiding spurious upward or downward trends or overfitting or underfitting the data (Paine et al., 2012).

Orthogonal polynomials are particularly popular for fitting biological curves because they have much lower correlations among their coefficients and provide estimates of the covariance matrices that tend to be more robust over a variety of data sets (Schaeffer, 2004). Legendre polynomials represent simple orthogonal polynomials and have been used successfully to fit longitudinal traits in livestock breeding programs (e.g., Albuquerque and Meyer, 2001; Oliveira et al., 2017, 2019b; Brito et al., 2017) and for plant research (e.g., Yang et al., 2006; Yang and Xu, 2007; Campbell et al., 2018; Momen et al., 2019).

Spline functions offer a more flexible alternative for modeling longitudinal traits compared to orthogonal polynomials (van Eeuwijk et al., 2018). Splines are piecewise polynomial functions, linked at specific points called knots (de Boor, 1980). For longitudinal data, these knots represent time points within the data collection interval (Li and Sillanpää, 2015). The greater flexibility of splines is due to the independence of each segment, which can have the same or different polynomial degrees, accommodating abrupt changes in the trajectory (Meyer, 2005b). A particular type of spline function is the basis spline, or B-spline (de Boor, 1980), extensively deployed in animal breeding (Meyer, 2005b; Oliveira et al., 2019a). Another version of spline is P-spline, which combines B-splines with different penalties on the coefficients of adjacent B-splines, resulting in smoother curves (Eilers and Marx, 1996; Meyer, 2005b).

Spline functions have recently been used to model longitudinal traits in crops. For instance, haulm senescence was assessed at several points during the growing season in a diploid potato mapping population and fitted using P-splines (Hurtado et al., 2012). Montesinos-López et al. (2017) used a B-spline function to fit wheat canopy hyperspectral bands in a yield prediction model. B-splines have also modeled rice temporal shoot biomass in a water-limited environment (Momen et al., 2019).

Statistical Genetic Models

Plant breeding is mostly based on the selection of new genetically superior cultivars from a large set of candidates. Simple arithmetic means of the phenotypic values, or Best Linear Unbiased Estimation (BLUE, treating genotypes as fixed effects), were used for selection prior to the development of Best Linear Unbiased Prediction (BLUP; Henderson, 1974). BLUPs are based on a mixed linear model and are now the most commonly used method for genetic evaluation of plant and livestock species (Piepho et al., 2008; Mrode, 2014). In the mixed-model framework, the genotypes are fitted as random and the genotypic effects are estimated by BLUP. The main advantage of BLUP over previous methods is its increased prediction accuracy for genetic effects. This is due to the shrinkage toward the mean that depends on the amount of information available (from the individual and/or its relatives), which will adjust extreme high and low performance toward the overall mean, and also to the incorporation of the genetic correlation between related genotypes from pedigree or genomic information (Piepho et al., 2008). The latter is not a requirement for the model, so the simplest case of BLUP uses no relationship matrix and the genotypes are considered to be independent random variables (Yan and Rajcan, 2003; Cullis et al., 2006). Piepho et al. (2008) present several examples of BLUP analyses in plant breeding.

Although rarely used, pedigree data is an easy and inexpensive source of information for plant breeders to leverage the relationship between individuals for a more accurate estimation of breeding values. Pedigree-based BLUPs have been successfully used in various crops (Bromley et al., 2000; Rutkoski et al., 2016; Basnet et al., 2019; Moreira et al., 2019), and have contributed to major advancements in the rates of genetic progress.

The inclusion of genomic information provides more accurate estimates of genetic relatedness among genotypes, especially with regards to the Mendelian sampling effects (Habier et al., 2007). Genomic information traces allele inheritance, capturing small segments of the genome shared among individuals, even when they are apparently unrelated through pedigree (Velazco et al., 2019). Plant breeders have widely adopted genomic-based BLUPs (GBLUP) for genomic selection (Auinger et al., 2016; Crossa et al., 2017; Schrag et al., 2019). Although genomic information is promising, in practice high-density genotyping is not always feasible for all genotypes within a breeding program due to genotyping costs, logistics, or both (Habier et al., 2009). An alternative is to construct a joint relationship matrix based on pedigree and genomic relationships to predict BLUPs for genotyped and non-genotyped material, which is called single-step GBLUP (ssGBLUP; Misztal et al., 2009; Aguilar et al., 2010; Christensen and Lund, 2010). This approach integrates both relationship matrices, connecting their different yet complementary information on genetic relatedness, and provides more reliable and accurate estimates of genetic similarities between genotypes. Genomic breeding values based on the ssGBLUP approach are commonly used in animal breeding (Aguilar et al., 2010; Legarra et al., 2014; Meuwissen et al., 2015; Guarini et al., 2019a, b; Oliveira et al., 2019c), and their use has started to become popular in plant breeding as well (Ashraf et al., 2016; Cappa et al., 2019; Velazco et al., 2019). In sorghum, Velazco et al. (2019) demonstrated that this methodology improves the predictive ability for complex traits, especially for traits with low heritability estimates, measured late in the development stage, or those that are difficult or expensive to measure.

For longitudinal traits, one can calculate BLUPs for each time point separately as individual traits with unique phenotypes; however, these approaches do not directly investigate and compare trends over time (Littell et al., 1998). This makes it difficult to consider a large number of time points and inhibit data comparison when BLUPs shrink differently due to discrepancies in heritability estimates. The main goal of fitting curves and patterns for longitudinal traits is to consider variability in the developmental process across many points in time (e.g., growth). Analytical methods have been developed to better evaluate longitudinal traits using the BLUP context, a simple analysis of variance, or both (Littell, 1990; Meyer and Kirkpatrick, 2005; Mrode, 2014). We will discuss the main methods in this review.

Repeatability Model

Individual measurements recorded over time can be treated as repeated records of the same trait. This is known as the repeatability model. There are two critical assumptions implicit in this method: (1) the variances of different measurements within the same genotype (or individual) are always equal, regardless of the time interval between records; and (2) the genetic correlations between all measurements are equal to one, i.e., measurements at different time points are all influenced by the same genes (Falconer and Mackay, 1996; Meyer and Hill, 1997; Littell et al., 1998). In this scenario, simple repeatability models are the standard approach.

One of the simplest methods is the repeated-measurements analysis of variance (ANOVA) using a split-plot in time design, which treats the genotypes as a whole-plot unit and genotypes at particular times as a sub-plot unit (Rowell and Walters, 1976; Littell, 1990). It is important to mention that as time is a factor in the experiment that cannot be randomized, this is not a true split-plot design. Also, this method assumes the data have equal variances (homoscedasticity) in all repeated measurements and that all pairs of measurements will have the same correlation (i.e., compound symmetry), which are unrealistic assumptions for most crop datasets. However, Huynh and Feldt (1970) showed that the equality of the variances of differences between any two treatment measurements assumed to be correlated was sufficient to perform a split-plot ANOVA. In this case, if the data violate the Huynh and Feldt condition, the F-statistics for the sub-plot unit and their interaction will be inflated. Thus, this method is prone to high Type I error rates, leading to conclusions that effects are statistically significant when they are not (Scheiner and Gurevitch, 2001; Fernandez, 2019).

In the context of mixed models, specifying the random and fixed effects in the model will depend on the study objectives, data structure, and the assumptions that can be made. Usually, time is considered as a fixed effect because it is not randomized in an experiment. Simple repeatability models have been used to calculate BLUPs and BLUEs of longitudinal traits derived from HTP for genomic prediction, such as in wheat (Rutkoski et al., 2016; Sun et al., 2017).

Multiple-Trait Model

Often, HTP platforms are used to generate phenotypes of plants in different “ages” or development stages, with the mean and variance of the phenotypes between measurements/assay dates changing over time. Thus, the assumption is that the genetic control of longitudinal traits will be different over time, characterizing the longitudinal records/phenotypes as different traits. A common approach to analyze longitudinal traits in this scenario is a multi-trait analysis that considers each time point as a different dependent variable (Sun et al., 2017).

Multivariate analysis of variance (MANOVA) is an extension of ANOVA, mentioned earlier, that avoids the covariance structure problems raised in repeated-measures ANOVA. However, it still requires equality in covariance among the groups being compared and balanced data over time. In addition, MANOVA assumes a multivariate normal distribution. Alternative methods have been proposed to overcome these restrictions (Krishnamoorthy and Lu, 2010; Krishnamoorthy and Yu, 2012; Konietschke et al., 2015), but MANOVA still has limited use in practice.

In the BLUP context, multiple-trait mixed models were first implemented by Henderson and Quaas (1976) to analyze two or more correlated traits making use of genetic and residual covariances among the traits (Speidel, 2011). Using this method, one can directly model the covariance structure of multiple dependent variables and efficiently handle missing data (Mrode, 2014). The main advantage of using a multiple-trait model (MTM) over a single-trait repeatability model is the improved evaluation accuracy for each trait arising from better connections in the data between the genetic and residual covariance (Colleau et al., 1999; Mrode, 2014). This data structure will benefit the prediction of traits with lower heritabilities when combined with highly heritable traits and genotypes with missing records for one or more traits (Mrode, 2014). In wheat, MTM was used to predict BLUPs for canopy temperature and normalized difference vegetation index (NDVI; Sun et al., 2017), and BLUEs for green NDVI (Juliana et al., 2018).

There are some disadvantages of multiple-trait mixed models. For instance, high-dimensional longitudinal data (e.g., traits recorded multiple times over a long period) can lead to over-parameterized models with high computational requirements (Speidel, 2011). There is also the potential for high correlations between consecutive measurements, which can reduce the power of the tests of significance (Foster et al., 2006). There are approaches to reduce the dimensionality of the MTM, which we are discussed below. It is worth noting that, when applying these approaches, the appropriate models should still be adequate to describe the data, accounting for the changes of mean and covariance over time, and estimate the necessary genetic parameters (Mrode, 2014).

Canonical transformation of phenotypic records is a common procedure to eliminate autocorrelation among traits through eigenvalue decomposition (Meyer and Hill, 1997). A set of highly correlated measures will provide eigenvalues close to zero. Under the framework of canonical transformed phenotypes, the original observations are transformed into a new set of response variables and the ones with the highest eigenvalues are selected to compose the new combination of traits. After fitting the MTM with the new values, the results are transformed back to the original scale (Mrode, 2014). Grosu et al. (2013) highlighted that canonical transformation can only be used if all individuals are recorded for all the traits and that the model needs to be the same for each trait, accommodating only two random effects: residual and genetic. Another strategy to fit MTM is referred to as “bending” (Thompson and Meyer, 1986; Meyer, 2019). It does not require all traits to be measured in all individuals. This procedure forces a decreased autocorrelation among traits by shrinking the covariance among traits by a bending factor, which creates a positive-definite covariance matrix.

The principal component analysis (PCA) and factor analysis (FA) methods are often more appropriate to reduce dimensionality for a large number of traits. FA identifies common factors, called latent variables, associated with the correlations between variables (Mrode, 2014). On the other hand, the PCA approach aims to create independent variables (principal components) that explain the maximum amount of variation in the dataset (Mrode, 2014). Thereafter, the principal components or latent variables become the new dependent variables in the MTM. Both methods have been used to reduce the dimensionality of longitudinal trait analysis in animals (Macciotta et al., 2017; Durón-Benítez et al., 2018; Vargas et al., 2018), and plants (Kwak et al., 2016; Yano et al., 2019).

As longitudinal traits are, by definition, taken along a time trajectory, the whole data set can be represented by parameters describing the shape of the trajectory curve (e.g., growth curve). These parameters can become the new dependent variables or integrated covariance structures in MTM (Speidel, 2011; Oliveira et al., 2019a); however, none of the approaches to analyze longitudinal data that we have discussed so far have considered that the genetic and environmental variances may change over time (Meyer, 1998, 2005a; Oliveira et al., 2019a). In addition, these approaches are limited to the time points at which traits were measured. Random regression models (RRMs) provide a way to overcome these limitations (Schaeffer, 2004).

Random Regression Model

A common property of longitudinal traits is that the covariance between repeated measures depends on the interval of time between them. In other words, measurements collected closer in time will be more correlated than measures collected farther apart. Kirkpatrick et al. (1990) presented the concept of analyzing longitudinal data using covariance functions by describing the covariance structure of the traits as a function of time. In essence, this approach fits a set of orthogonal functions to a given covariance matrix for the records taken over time (Meyer and Hill, 1997).

First-order autoregressive analysis (AR-1) is an appealing method for modeling covariance structure for phenotypes measured over time (Apiolaza and Garrick, 2001; Yang et al., 2006; Vanhatalo et al., 2019). It assumes homogenous variances and correlations that decline exponentially as measurements are separated by greater time intervals. Thus, two measurements collected closer in time will be more correlated than those further apart (Wade et al., 1993; Littell et al., 2000; Piepho et al., 2004). The AR-1 structure is only applicable for measurements taken at equally spaced time points (Wang and Goonewardene, 2004). Though this is a difficult requirement to meet in agricultural research, especially in field trials, modeling the longitudinal trait as described in the previous section would make the data evenly spaced over time and validate the AR-1 method. An alternative is to use a spatial power covariance structure that allows for unequal intervals between time points (Wang and Goonewardene, 2004).

Note that so far, we are assuming homogenous variance over time. There are also covariance structures to handle heterogeneous variance, such as the first-order ante-dependence structure (Wolfinger, 1996). Thus, Legendre orthogonal polynomials and splines are more attractive covariance functions as they produce relatively small correlations among the regression parameters and adjust flexibly to the shape of the trajectory curve (Schaeffer, 2004; Meyer, 2005a, b; Bohmanova et al., 2008; Pereira et al., 2013; Brito et al., 2018). In plants, different covariance structures have been assessed for a variety of traits (Apiolaza et al., 2011; Sun et al., 2017; Campbell et al., 2019).

Meyer and Hill (1997) showed that covariance functions are equivalent to RRMs. Schaeffer (2016) reported that covariance functions help to predict the change in variation over time, while RRMs are a way to estimate covariance functions and determine individual differences in trajectories. RRMs provide a robust framework for modeling trait trajectories using covariance at or between each time point with no assumptions of constant variances or correlations. RRMs provide insights about the temporal genetic variation of developmental behavior underlying the studied traits (Oliveira et al., 2019a). Despite the increased computational cost, RRMs result in more accurate breeding values compared to other methods (Sun et al., 2017; Oliveira et al., 2019a).

The RRMs were first introduced in animal breeding to overcome over-parameterized models in MTM and they have been used extensively since then (Jamrozik and Schaeffer, 1997; Schaeffer, 2004; van Pelt et al., 2015; Englishby et al., 2016; Oliveira et al., 2019a). In summary, RRMs set the parameters of the function describing the trajectory of the trait as fixed and random effects in the model, resulting in fewer parameters than MTM (Schaeffer, 2016; Oliveira et al., 2019a). Consequently, in RRMs the random parameters do not correspond directly to the individuals’ genetic value for the longitudinal trait. Rather, they correspond to the genetic values of sets of regression coefficients that represent the time trajectory of the longitudinal trait for each genotype (Turra et al., 2012). Estimates of genetic parameters and breeding values can be obtained for all time points within the interval analyzed from the genetic (co)variance matrices for the regression coefficients and the matrix of independent covariates for all time points associated with the function used (Oliveira et al., 2019a). When the same fixed effects are used in all models, it is appropriate to examine different covariance structures using real data and select the one that best fits the model based on a statistical methods, such as the Akaike Information Criterion (AIC, Wang and Goonewardene, 2004) or Bayesian Information Criterion (BIC, Neath and Cavanaugh, 2012). Finally, estimate the effects of interest using the selected covariance structure. In a general form, RRM can be described as follows (Oliveira et al., 2019a):

where y_gij is the jth repeated record of genotype i (e.g., canopy coverage at different days after planting); b_qg is the qth fixed regression coefficient for the gth group; a_ri is the rth random regression coefficient for the additive genetic effect of the ith genotype; p_si is the sth random regression coefficient for permanent environmental effect of the ith genotype; e_gij is the residual effect; and z_qg, z_ri and z_si are the covariates related to the function used to describe time (e.g., days after planting), assuming the same function (e.g., Legendre polynomial) with possible different orders Q, R, and S (e.g., linear, quadratic, cubic) (Oliveira et al., 2019a).

Random regression models have been shown to be the most effective choice to genetically evaluate longitudinal traits in numerous livestock breeding programs (as reviewed by Oliveira et al., 2019a), but there are only a few examples of the applications of RRMs in plant breeding, especially when incorporating genomic information. Sun et al. (2017) captured the change of HTP traits continually over wheat growth stages using RRMs. Campbell et al. (2018) used RRMs to predict shoot growth trajectories in a rice diversity panel and demonstrated an improvement in prediction accuracy compared to a single-time-point model. Based on the same rice dataset, Campbell et al. (2019) used RRMs to identify QTL with time-specific effects. Multiple-trait RRMs are also feasible and have been implemented in several settings in animal breeding programs (Nobre et al., 2003; Muir et al., 2007; Oliveira et al., 2019b, c).

Implementation of Genomic Selection for Longitudinal Traits

Meuwissen et al. (2001) introduced the concept of genomic selection (GS) based on the idea that markers from dense genome-wide genotyping will be in linkage disequilibrium with QTLs that have an effect on the quantitative trait of interest. Thus, they can be used for selection without identifying the QTL or the functional polymorphism. This increased understanding of GS arose as it became known that markers would carry relationship information in addition to the signal captured by the linkage disequilibrium between markers and QTL (Habier et al., 2007; Meuwissen, 2009).

In GS, genomic and phenotypic data are combined in a training population to enable the development of prediction equations that can be used in a testing (or target) population of selection candidates consisting of individuals that were genotyped but not phenotyped (Crossa et al., 2017). Therefore, GS enables a more accurate selection of individuals at an early age (with no measurements). This increases the rate of genetic gain by reducing the time required for the variety development and the cost per cycle. HTP is able to generate high-quality quantitative data and effectively characterizes large training populations during the growing season. The combination of GS and HTP has the potential to increase precision and efficiency while lowering costs and minimizing labor (Araus et al., 2018).

Under the longitudinal framework of GS, the prediction of temporal breeding values enables targeted selection on specific periods in the growing season or selection of individuals that exhibit desirable trait trajectories. In addition, the longitudinal trait can be used as secondary traits to improve the genomic selection of economic endpoint traits such as yields (Sun et al., 2017). Campbell et al. (2018) used RRMs with a second-order Legendre polynomial to perform pedigree and genomic predictions of shoot growth trajectories in a rice diversity panel. They demonstrated an improvement in prediction accuracy using the RRM compared to a single-time-point model. Furthermore, the authors reported that genomic RRMs were useful in predicting future phenotypes using a subset of early measurements. Another study in rice used RRMs to predict projected shoot area in controlled and water-limited conditions using Legendre polynomials and B-spline basis functions (Momen et al., 2019). Before fitting both functions, they adjusted raw phenotypic measurements to obtain BLUEs for downstream genetic analysis. Overall, RRMs produced higher prediction accuracy compared to the baseline multiple-trait model. In addition, B-splines performed slightly better than Legendre polynomials (Momen et al., 2019).

Currently, statistical models used in GS for plant breeding are most often single-trait (univariate) and do not take advantage of genetic covariance among traits or phenotypic records collected at different time points (Jia et al., 2012). However, MTM for GS was shown to outperform single-trait models by accounting for correlation among traits, thereby increasing prediction accuracy, statistical power, parameter estimation accuracy, and reducing trait selection bias (Jia et al., 2012; Guo G. et al., 2014; Montesinos-López et al., 2016, 2019). These advantages are even more obvious for low-heritability traits, such as yield, that are genetically correlated with highly heritable traits (Guo G. et al., 2014; Jiang et al., 2015). Recently, studies in the CIMMYT (2019) wheat breeding program¹ have shown that the accuracy of GS is greatly improved by incorporating HTP longitudinal data from the so-called secondary traits measured with UAV (Rutkoski et al., 2016; Montesinos-López et al., 2017; Sun et al., 2017, 2019), an approach that is relatively inexpensive to implement as HTP and genotyping have become more accessible (e.g., targeted genotyping-by-sequencing approach; Pembleton et al., 2016). In addition, secondary traits are also useful to predict the primary trait at early growth stages, since they can often be phenotyped ahead of a primary trait like grain yield (Sun et al., 2017). Therefore, longitudinal traits can be used as secondary traits to improve the accuracy of GS and contribute to a better understanding of the biological mechanisms underlying stress responses and development. As described in the previous section, there are various ways to extract the genetic information from longitudinal traits and the methods employed will determine how they can be used in GS.

Rutkoski et al. (2016) used HTP canopy temperature (CT), green normalized difference vegetation index (GNDVI), and red normalized difference vegetation index (RNDVI) taken over time as secondary traits in GS for yield in wheat. First, they estimated BLUEs for the longitudinal traits using the repeatability model and used them in an MTM with yield, for pedigree and genomic predictions. They found that multiple-trait modeling with secondary traits increased accuracies for grain yield using both pedigree and genomic information, compared to the single-trait models. In another study, CT and NDVI also improved the ability to predict grain yield in wheat (Sun et al., 2017). However, in addition to a repeatability model, the authors also used MTM and RRMs to calculate BLUPs for the secondary traits in order to compare their efficiency. The predictive ability improved by 70%, on average, when including secondary traits, and the predictive ability of RRM and MTM were superior to the repeatability model. Also in wheat, Juliana et al. (2018) performed pedigree and genomic multi-trait prediction models using BLUEs of yield and GNDVI measured at different dates. They found that including GNDVI increased prediction accuracies. Sun et al. (2019) used an RRM with a cubic smoothing spline to predict BLUPs for CT and GNDVI in wheat. In a second step, they used BLUPs for secondary traits and grain yield as the dependent variables in GS. The prediction accuracy using the secondary traits increased by an average of 146% for grain yield across cycles and the secondary traits measured in the early stages were optimal for enhancing the prediction accuracy. Montesinos-López et al. (2017) and Crain et al. (2018) obtained similar results in wheat. Howard and Jarquin (2019) modeled the genetic covariance between canopy coverage and yield using the SoyNAM dataset (Song et al., 2017; Diers et al., 2018) and demonstrated that, based on different cross-validation schemes, the predictive ability was the highest when both canopy and marker information were included in the model. Two other papers reported similar improvements with the same dataset (Xavier et al., 2017; Jarquin et al., 2018).

Given the capability of HTP to collect multiple temporal traits at the same time, multiple-trait RRMs (MTRRMs) can be powerful tools for joint genomic prediction of several longitudinal traits (Oliveira et al., 2016). In addition, MTRRMs can incorporate different functions to describe different traits in the same model and estimate genetic correlations between different traits over time (Oliveira et al., 2016). In animals, MTRRMs are a plausible alternative for joint genetic prediction of milk yield and milk constituents in goats (Oliveira et al., 2016), cattle (Oliveira et al., 2019c), and buffaloes (Borquis et al., 2013). Recently, MTRRMs for projected shoot area and water-use recorded daily over a period of 20 days showed better predictive abilities compared to single-trait RRMs (Baba et al., 2020).

In animal breeding and multiple-stage plant breeding analysis, it is common to use deregressed genetic values as the pseudo-phenotypes for genomic predictions. Oliveira et al. (2018) compared different deregression methods for longitudinal traits. However, this multiple-step approach may result in lower accuracies, bias, and loss of information (Legarra et al., 2009; Kang et al., 2017). Considering the advantages of ssGBLUP and RRMs in genetic evaluation, integrating both approaches is an effective strategy to enhance the genomic prediction of longitudinal traits (Kang et al., 2017). Koivula et al. (2015) reported higher accuracy and less bias in the prediction of Nordic Red Dairy cows for milking performance using a ssGBLUP RRM compared to the traditional pedigree-based RRM. Kang et al. (2017) showed that ssGBLUP RRMs achieved the highest accuracy and least bias under a variety of scenarios, including persistency of accuracy over generations, compared to other models. In summary, the use of ssGBLUP based on RRMs can increase the reliability of genomic predictions for test-day traits in dairy cattle (Koivula et al., 2015; Kang et al., 2018; Oliveira et al., 2019c), and possibly in crops.

Detecting QTL and Causal Variants Associated With Longitudinal Traits

One of the main goals in genomic research is to predict the phenotypic variation using genotypes, by identifying genetic variants. The development of an organism is the result of an interacting network of genes and environmental factors (Wu and Lin, 2006). Unlike single-time-point measurements, studying longitudinal traits as a function of time allows the comprehensive assessment of crop growth and development (e.g., age metabolic rate; Ma et al., 2002). However, in plants, the detection of QTL analysis or genome-wide association studies (GWAS) for longitudinal traits are still performed at each time point independently. For instance, Würschum et al. (2014) used linkage mapping at discrete time points separately to identify time-specific QTLs associated with plant height in triticale. In cotton, canopy-related traits were used separately for each of the several studied days to map additive QTL effects and their interaction with the environment (Pauli et al., 2016a). In soybeans, a GWAS was used to identify QTL for each individual canopy coverage measurement spanning 14–56 days after planting (Xavier et al., 2017). Zhang et al. (2017) performed QTL mapping for several growth-related traits at 16 time points separately in maize. Also in maize, analysis of individual time points found different and simultaneous QTLs controlling plant height at different growth stages (Wang et al., 2019). Using time point growth-related traits, Knoch et al. (2020) found evidence of temporal QTLs in canola. Although useful, these static examinations provide a simplified view of genetic control, neglecting temporal changes and developmental features of trait formation (Wu and Lin, 2006). In addition, in animals, it has been shown that neither the phenotypic nor additive polygenic effects of longitudinal traits are constant throughout the entire phenotypic expression (Szyda et al., 2014; Brito et al., 2018; Oliveira et al., 2019a).

As an alternative, Ma et al. (2002) proposed a dynamic model, called functional mapping, to map QTLs associated with the whole developmental process of longitudinal traits. As mentioned earlier, longitudinal traits can be represented as curves, described by a few parameters from a linear or non-linear function over a given time. The idea behind functional mapping is that the difference in curve parameters among genotypes may suggest temporal patterns of genetic control over the phenotypic trajectory (Ma et al., 2002). Thus, functional mapping allows testing of timing and the duration of QTL expression (Wu et al., 2004). Several modeling strategies for functional mapping have been proposed and have been reviewed by Li and Sillanpää (2015). One of the approaches (the two-stage method) consists of modeling the whole phenotypic trajectory using linear and non-linear models and using these parameters as latent-trait phenotypes for QTL detection (Li and Sillanpää, 2015). Often, researchers perform analyses for individual time points followed by this two-stage method to derive the curve parameters. Busemeyer et al. (2013) used a logistic function to fit high-throughput-derived biomass from different development stages of a large mapping population of 647 double-haploid triticale lines. In addition to GWAS for the individual days, they performed a multiple-trait functional GWAS using the parameters from the logistic curve to reveal temporal genetic patterns of biomass regulation. A similar approach was used to assess image-derived biovolume in maize lines (Muraya et al., 2017); digital biomass accumulation in spring barley (Neumann et al., 2017); and area, height, and senescence in wheat (Camargo et al., 2018). Campbell et al. (2017) calculated the projected shoot area in 360 rice accessions from 19 to 41 days after transplanting. They modeled the longitudinal phenotypes using a power function (Paine et al., 2012) and used the parameters as the pseudo-phenotypes in a multiple-trait GWAS. In order to reveal the temporal dynamics of senescence in potato, Hurtado et al. (2012) employed P-splines as a smoothing curve and used the curve parameters for identifying QTLs.

Kwak et al. (2014) proposed two simple regression-based methods to map QTL by analyzing each time point separately and then combining test statistics across time points to determine the overall significance. Later, Kwak et al. (2016) proposed an improved approach where the observed longitudinal traits were replaced by a smoothing approximation, followed by dimensional reduction via PCA. Multiple-trait QTL analysis was then performed on the reduced data (using principal components). Muraya et al. (2017) implemented the approach suggested by Kwak et al. (2016). They used B-splines to smooth the phenotype followed by PCA for variable reduction and performed a multiple−QTL analysis following the methods of Kwak et al. (2014) to reveal the underlying genetic variation of growth dynamics in maize. Temporal height QTLs in the model C4 grass Setaria were also revealed using this approach (Feldman et al., 2017). Animal breeders have been using PCA for longitudinal trait analysis for a long time to synthesize complex patterns and reduce computationally demanding multiple-trait QTL detection (Macciotta et al., 2006, 2015, 2017; Zhang et al., 2018).

RRMs offer a better option to fit longitudinal traits and have been widely used in genetic evaluation of animals (Ning et al., 2017; Oliveira et al., 2019a). The random regression approach uncovers SNP effects over time because it is able to identify persistent and time-specific transient QTLs. Moreover, RRMs have increased statistical power to detect QTLs over other approaches because they leverage the full set of raw longitudinal phenotypes (Ning et al., 2017) and can capture QTLs with significant effects in specific regions of the development curve, though the effects of these QTLs may be small overall. RRMs are also useful for detecting QTL in gene-by-environment interactions (Lillehammer et al., 2007; Carvalheiro et al., 2019).

Das et al. (2011) proposed a method called functional GWAS (fGWAS), based on RRMs, which integrates GWAS and mathematical models describing biological processes. In summary, fGWAS estimates the mean for different SNP effects for each genotype and time point and then performs hypothesis testing to determine whether the SNP has any additive or dominant effect during the time course. The main drawback of this method is that it only performs a single-locus analysis. Later, Ning et al. (2017) proposed a modification of fGWAS by estimating the time-dependent population mean and the SNP effects separately, instead of fitting them directly. They also extended the model to capture the time-varying polygenic effect of complex traits by treating SNPs as covariates (fGWAS-C) or factors (fGWAS-F). However, their method was shown to be computationally inefficient due to the high dimensionality of the mixed model equations compared to other models. Subsequently, Ning et al. (2018) proposed a rapid longitudinal GWAS method, transforming the covariance matrices to diagonal matrices using eigen-decomposition. This way the model can be solved by a weighted least squares model for each SNP test.

To the best of our knowledge, Campbell et al. (2019) were the first to use RRM GWAS in a major row crop. They took the genomic breeding values derived from RRMs using Legendre orthogonal polynomials to assess the genetic architecture of rice shoot growth over a period of 20 days during early vegetative growth. They found both transient and persistent effects associated with shoot growth and more associations with the RRM when compared to single-time-point analysis.

Challenges and Future Developments

To capitalize on advances in phenotyping and molecular technologies, greater progress is needed in developing ways in which breeders can manipulate systems to understand the relationships between genotype and phenotype. The underlying biological changes due to environmental systems and/or over time can be captured with longitudinal data. The major challenge is synthesizing the various layers of information together in a meaningful manner to understand the downstream effects of developmental stress and implications for breeding (Harfouche et al., 2019).

Non-Additive Effects and GxE

Non-additive genetic effects may significantly contribute to the total genetic variation of complex traits. Prediction models that include dominance effects represent an important component of breeding programs that focus on crossbred populations, hybrid production, and vegetatively-propagated species (de Almeida Filho et al., 2019). There is also ample evidence of the importance of epistasis in the genetic architecture of complex traits in various crops (Guo T. et al., 2014; Monir and Zhu, 2018). Integrating non-additive effects into statistical models may improve prediction accuracy and detect more QTLs than simple additive models, especially when the non-additive variance contributes to a large proportion of the genetic variance (Bouvet et al., 2016; Bonnafous et al., 2018; Liu et al., 2018b, 2019; Monir and Zhu, 2018; Varona et al., 2018). However, these studies are restricted to single-time-point traits. For longitudinal traits, it may be challenging to have a full genetic model (including both additive and non-additive effects), requiring dense marker panels to estimate the time-dependent (co)variances, as well as partitioning of the genetic variance components. Nevertheless, full genetic models of longitudinal traits may have the potential to impact future design and implementation of breeding strategies.

The temporal dynamics of longitudinal traits lead to interactions that change the phenotype over time. This may be because the gene-gene and gene-environment interactions (GxE) are time- or age-dependent and need to be properly modeled (Fan et al., 2012). In this case, environmental descriptors should be measured several times as the trait phenotypes. The resulting model is a multiple-trait, multiple-environment model with a variety of interactions, in which computational issues may arise due to the increase in the number of parameters being estimated. It has been shown that RRMs can account simultaneously for the additive genetic effect and some degree of GxE in longitudinal traits in animal breeding by allowing for the estimation of genetic (co)variance components and breeding values over the whole trajectory of a time-dependent trait and environment-dependent covariate (Brügemann et al., 2011; Santana et al., 2016; Bohlouli et al., 2019). In plant breeding, therefore, this model may provide considerable biological insights into the mechanisms determining performance in specific environments, making it a worthwhile method for study in future research.

Complementary “-Omics” Technologies

The rapid advances in “-omics” technologies enable the generation of large-scale “-omics” datasets for many crop species, providing new opportunities to investigate and improve complex traits. The different approaches described in this review offer valuable tools to combine phenomics and genomics data to reveal the underlying genetic basis of longitudinal traits. However, one current challenge is integrating additional “-omics” technologies (e.g., transcriptomics, metagenomics, proteomics, metabolomics, epigenomics) to provide a holistic multi-omics approach to study biological mechanisms and their response to environmental stresses for important agronomic traits.

Recently, methods that combine “-omics” information have been used in some crops to study phenotypic networks for single-point traits, for example, to pinpoint candidate genes and/or loci and predict phenotypic variation (Acharjee et al., 2016; Li et al., 2016; Das et al., 2017; Sheng et al., 2017; Pandey et al., 2018; Jiang et al., 2019). A meta-analysis of the detailed “-omics” datasets regarding longitudinal traits in crops has been limited so far. Baker et al. (2019) characterized the mechanistic connections between the genomic architecture, transcriptomic expression networks, and phenotypic variation of growth curves that underlie the developmental dynamics of plant height in Brassica rapa. The combination of multi-omics approaches also seems promising to elucidate senescence processes in model and crop plants (Großkinsky et al., 2018). When joint modeling longitudinal “-omics” data (one or more type of “-omics” data measured over time) the statistical analysis becomes more challenging. Some key points can be found in Sperisen et al. (2015). In general, there is a need to adapt methodologies and experimental designs to explore processes related to the global evolution of biological processes such as growth and development. Despite all these challenges, integrative methods can increase analysis power to find true causal variants, regulatory networks, and pathways. These, in turn, could be incorporated into GS and breeding programs to speed up genetic gains (Suravajhala et al., 2016).

Deep Learning

Deep learning (DL) is a powerful and highly flexible class of machine learning algorithms based on representation-learning methods that incorporates multiple levels in a non-linear hierarchical learner (Lecun et al., 2015). Essentially, DL is an advanced version of artificial neural networks (ANN) with multiple hidden layers that aims to mimic the human brain functioning (Patterson and Gibson, 2017).

Deep learning has demonstrated its utility in different fields of biological sciences, such as disease diagnosis (Gao et al., 2018), multi-omics data integration (Chaudhary et al., 2018), predicting DNA- and RNA-binding specificity (Trabelsi et al., 2019), and, recently, in plant breeding genomic prediction (Ma et al., 2018; Montesinos-López A. et al., 2018; Montesinos-López O. A. et al., 2018; Montesinos-López et al., 2019). Zou et al. (2019) and Pérez-Enciso and Zingaretti (2019) provide a primer on DL in genomics. The growing interest in DL methods in plant breeding, especially for prediction, may arise from its powerful capability of learning complex non-linear relationships between predictors and responses hidden in big data, usually resulting in higher accuracy when compared with other methods (Montesinos-López A. et al., 2018; Pérez-Enciso and Zingaretti, 2019; Zou et al., 2019). It is important to point out that even though DL can deal with complex scenarios and achieve state-of-the-art accuracy, it requires domain knowledge and large-scale datasets, while the interpretation of the underlying biology is more challenging than for standard statistical models (Zou et al., 2019).

Within the classes of DL, recurrent neural networks (RNN) are designed for sequential or time-series data (Lecun et al., 2015) and may be the most appropriate architecture to model longitudinal traits. An RNN can be thought of as a memory state that retains information on previous data the network has seen and updates its predictions in the light of new information. Thus, besides prediction, RNN has the ability to capture long-term temporal dependencies (Che et al., 2018). Recently, RNNs have achieved astonishing results in many applications with time series or sequential data, particularly in human sciences (Azizi et al., 2018; Che et al., 2018; Lee et al., 2019; Sung et al., 2019; Zhong et al., 2019). Despite its advantages, to our best knowledge RNNs have not been employed in genomic prediction or mapping QTL for longitudinal traits in plant breeding. In the context, versatile DL models for multiple-trait analysis (Montesinos-López O. A. et al., 2018), multiple-environment analysis (Montesinos-López O. A. et al., 2018), and performing simultaneous predictions of mixed phenotypes (binary, ordinal and continuous; Montesinos-López et al., 2019) have been successfully implemented. Such cases are not only encouraging but may lead to future integration of DL and RNNs into the analysis of longitudinal traits in crops. DL is a powerful approach and is likely to transform many domains in plant breeding because it has the potential to handle all the complexities highlighted in this review. Needless to say, further innovation and technology assessment are required to fully enable DL to deal with the unique properties of planting breeding data.

References

• 1

  AcharjeeA.KloostermanB.VisserR. G. F.MaliepaardC. (2016). Integration of multi-omics data for prediction of phenotypic traits using random forest.BMC Bioinformatics17:180. 10.1186/s12859-016-1043-4

  • CrossRef
  • Google Scholar

• 2

  AguilarI.MisztalI.JohnsonD. L.LegarraA.TsurutaS.LawlorT. J. (2010). Hot topic: a unified approach to utilize phenotypic, full pedigree, and genomic information for genetic evaluation of Holstein final score.J. Dairy Sci.93743–752. 10.3168/jds.2009-2730

  • CrossRef
  • Google Scholar

• 3

  AlbuquerqueL. G.MeyerK. (2001). Estimates of covariance functions for growth from birth to 630 days of age in Nelore cattle.J. Anim. Sci.792776–2789. 10.2527/2001.79112776x

  • CrossRef
  • Google Scholar

• 4

  AnN.PalmerC. M.BakerR. L.MarkelzR. J. C.TaJ.CovingtonM. F.et al (2016). Plant high-throughput phenotyping using photogrammetry and imaging techniques to measure leaf length and rosette area.Comput. Electron. Agric.127376–394. 10.1016/J.COMPAG.2016.04.002

  • CrossRef
  • Google Scholar

• 5

  ApiolazaL. A.GarrickD. J. (2001). Analysis of Longitudinal Data from Progeny Tests: Some Multivariate Approaches. Available at: https://academic.oup.com/forestscience/article-abstract/47/2/129/4617386(accessed June 4, 2019).

  • Google Scholar

• 6

  ApiolazaL. A.GilmourA. R.GarrickD. J. (2011). Variance modelling of longitudinal height data from a Pinus radiata progeny test.Can. J. For. Res.30645–654. 10.1139/x99-246

  • CrossRef
  • Google Scholar

• 7

  ArausJ. L.CairnsJ. E. (2014). Field high-throughput phenotyping: the new crop breeding frontier.Trends Plant Sci.1952–61. 10.1016/j.tplants.2013.09.008

  • CrossRef
  • Google Scholar

• 8

  ArausJ. L.KefauverS. C.Zaman-AllahM.OlsenM. S.CairnsJ. E. (2018). Translating High-Throughput Phenotyping into Genetic Gain.Amsterdam: Elsevier, 10.1016/j.tplants.2018.02.001

  • CrossRef
  • Google Scholar

• 9

  AshrafB.EdrissV.AkdemirD.AutriqueE.BonnettD.CrossaJ.et al (2016). Genomic prediction using phenotypes from pedigreed lines with no marker data.Crop Sci.56957–964. 10.2135/cropsci2015.02.0111

  • CrossRef
  • Google Scholar

• 10

  AtkinsonJ. A.PoundM. P.BennettM. J.WellsD. M. (2019). Uncovering the hidden half of plants using new advances in root phenotyping.Curr. Opin. Biotechnol.551–8. 10.1016/j.copbio.2018.06.002

  • CrossRef
  • Google Scholar

• 11

  AuingerH. J.SchönlebenM.LehermeierC.SchmidtM.KorzunV.GeigerH. H.et al (2016). Model training across multiple breeding cycles significantly improves genomic prediction accuracy in rye (Secale cereale L.).Theor. Appl. Genet.1292043–2053. 10.1007/s00122-016-2756-5

  • CrossRef
  • Google Scholar

• 12

  AziziS.BayatS.YanP.TahmasebiA.KwakJ. T.XuS.et al (2018). Deep recurrent neural networks for prostate cancer detection: analysis of temporal enhanced ultrasound.IEEE Trans. Med. Imaging372695–2703. 10.1109/TMI.2018.2849959

  • CrossRef
  • Google Scholar

• 13

  BabaT.MomenM.CampbellM. T.WaliaH.MorotaG. (2020). Multi-trait random regression models increase genomic prediction accuracy for a temporal physiological trait derived from high-throughput phenotyping.PLoS ONE15:e0228118. 10.1371/journal.pone.0228118

  • CrossRef
  • Google Scholar

• 14

  BaenzigerP. S.RussellW. K.GraefG. L.CampbellB. T. (2006). Improving lives: 50 Years of crop breeding, genetics, and cytology (C-1).Crop Sci.20132230–2244. 10.2135/cropsci2005.11.0404gas

  • CrossRef
  • Google Scholar

• 15

  BaillotN.GirousseC.AllardV.Piquet-PissalouxA.Le GouisJ. (2018). Different grain-filling rates explain grain-weight differences along the wheat ear.PLoS ONE13:e0209597. 10.1371/journal.pone.0209597

  • CrossRef
  • Google Scholar

• 16

  BakerR. L.LeongW. F.BrockM. T.RubinM. J.MarkelzR. J. C.WelchS.et al (2019). Integrating transcriptomic network reconstruction and eQTL analyses reveals mechanistic connections between genomic architecture and Brassica rapa development.PLoS Genet.15:e1008367. 10.1371/journal.pgen.1008367

  • CrossRef
  • Google Scholar

• 17

  BasnetB. R.CrossaJ.DreisigackerS.Pérez-RodríguezP.ManesY.SinghR. P.et al (2019). Hybrid wheat prediction using genomic, pedigree, and environmental covariables interaction models.Plant Genome121–13. 10.3835/plantgenome2018.07.0051

  • CrossRef
  • Google Scholar

• 18

  BlanconJ.DutartreD.TixierM.-H.WeissM.ComarA.PraudS.et al (2019). A high-throughput model-assisted method for phenotyping maize green leaf area index dynamics using unmanned aerial vehicle imagery.Front. Plant Sci.10:685. 10.3389/fpls.2019.00685

  • CrossRef
  • Google Scholar

• 19

  BodnerG.NakhforooshA.ArnoldT.LeitnerD. (2018). Hyperspectral imaging: a novel approach for plant root phenotyping.Plant Methods14:84. 10.1186/s13007-018-0352-1

  • CrossRef
  • Google Scholar

• 20

  BohlouliM.AlijaniS.NaderiS.YinT.KönigS. (2019). Prediction accuracies and genetic parameters for test-day traits from genomic and pedigree-based random regression models with or without heat stress interactions.J. Dairy Sci.102488–502. 10.3168/jds.2018-15329

  • CrossRef
  • Google Scholar

• 21

  BohmanovaJ.MigliorF.JamrozikJ.MisztalI.SullivanP. G. (2008). Comparison of random regression models with legendre polynomials and linear splines for production traits and somatic cell score of canadian holstein cows.J. Dairy Sci.913627–3638. 10.3168/jds.2007-0945

  • CrossRef
  • Google Scholar

• 22

  BonnafousF.FievetG.BlanchetN.BonifaceM.-C. C.CarrèreS.GouzyJ.et al (2018). Comparison of GWAS models to identify non-additive genetic control of flowering time in sunflower hybrids.Theor. Appl. Genet.131319–332. 10.1007/s00122-017-3003-4

  • CrossRef
  • Google Scholar

• 23

  BorquisR. R. A.NetoF. R.deA.BaldiF.Hurtado-LugoN.de CamargoG. M. F.et al (2013). Multiple-trait random regression models for the estimation of genetic parameters for milk, fat, and protein yield in buffaloes.J. Dairy Sci.965923–5932. 10.3168/JDS.2012-6023

  • CrossRef
  • Google Scholar

• 24

  BouvetJ. M.MakouanziG.CrosD.VigneronP. (2016). Modeling additive and non-additive effects in a hybrid population using genome-wide genotyping: prediction accuracy implications.Heredity (Edinb)116146–157. 10.1038/hdy.2015.78

  • CrossRef
  • Google Scholar

• 25

  BradshawJ. E. (2017). Plant Breeding: Past, Present and Future.Dordrecht: Springer, 10.1007/978-3-319-23285-0

  • CrossRef
  • Google Scholar

• 26

  BritoL. F.SilvaF. G.OliveiraH. R.SouzaN. O.CaetanoG. C.CostaE. V.et al (2017). Modelling lactation curves of dairy goats by fitting random regression models using Legendre polynomials or B-splines.Can. J. Anim. Sci.9873–83. 10.1139/cjas-2017-0019

  • CrossRef
  • Google Scholar

• 27

  BritoL. F.SilvaF. G.OliveiraH. R.SouzaN. O.CaetanoG. C.CostaE. V.et al (2018). Modelling lactation curves of dairy goats by fitting random regression models using Legendre polynomials or B-splines.Can. J. Anim. Sci.9873–83. 10.1139/cjas-2017-0019

  • CrossRef
  • Google Scholar

• 28

  BromleyC. M.Van VleckL. D.JohnsonB. E.SmithO. S. (2000). Estimation of genetic variance in corn from F1 performance with and without pedigree relationships among inbred lines.Crop Sci.40651–655. 10.2135/cropsci2000.403651x

  • CrossRef
  • Google Scholar

• 29

  BrügemannK.GernandE.von BorstelU. U.KönigS. (2011). Genetic analyses of protein yield in dairy cows applying random regression models with time-dependent and temperature x humidity-dependent covariates.J. Dairy Sci.944129–4139. 10.3168/jds.2010-4063

  • CrossRef
  • Google Scholar

• 30

  BurridgeJ.JochuaC. N.BuckschA.LynchJ. P. (2016). Legume shovelomics: high-throughput phenotyping of common bean (Phaseolus vulgaris L.) and cowpea (Vigna unguiculata subsp, unguiculata) root architecture in the field.F. Crop. Res.19221–32. 10.1016/j.fcr.2016.04.008

  • CrossRef
  • Google Scholar

• 31

  BusemeyerL.RuckelshausenA.MöllerK.MelchingerA. E.AlheitK. V.MaurerH. P.et al (2013). Precision phenotyping of biomass accumulation in triticale reveals temporal genetic patterns of regulation.Sci. Rep.3:2442. 10.1038/srep02442

  • CrossRef
  • Google Scholar

• 32

  CamargoA. V.MackayI.MottR.HanJ.DoonanJ. H.AskewK.et al (2018). Functional mapping of quantitative trait loci (QTLs) associated with plant performance in a wheat MAGIC mapping population.Front. Plant Sci.9:887. 10.3389/fpls.2018.00887

  • CrossRef
  • Google Scholar

• 33

  CampbellM.MomenM.WaliaH.MorotaG. (2019). Leveraging breeding values obtained from random regression models for genetic inference of longitudinal traits.Plant Genome12:435685. 10.3835/plantgenome2018.10.0075

  • CrossRef
  • Google Scholar

• 34

  CampbellM.WaliaH.MorotaG. (2018). Utilizing random regression models for genomic prediction of a longitudinal trait derived from high-throughput phenotyping.Plant Direct2:e00080. 10.1002/pld3.80

  • CrossRef
  • Google Scholar

• 35

  CampbellM. T.DuQ.LiuK.BrienC. J.BergerB.ZhangC.et al (2017). A comprehensive image-based phenomic analysis reveals the complex genetic architecture of shoot growth dynamics in rice (Oryza sativa).Plant Genome101–14. 10.3835/plantgenome2016.07.0064

  • CrossRef
  • Google Scholar

• 36

  CappaE. P.de LimaB. M.da Silva-JuniorO. B.GarciaC. C.MansfieldS. D.GrattapagliaD. (2019). Improving genomic prediction of growth and wood traits in Eucalyptus using phenotypes from non-genotyped trees by single-step GBLUP.Plant Sci.2849–15. 10.1016/j.plantsci.2019.03.017

  • CrossRef
  • Google Scholar

• 37

  CarvalheiroR.CostillaR.NevesH. H. R.AlbuquerqueL. G.MooreS.HayesB. J. (2019). Unraveling genetic sensitivity of beef cattle to environmental variation under tropical conditions.Genet. Sel. Evol.51:29. 10.1186/s12711-019-0470-x

  • CrossRef
  • Google Scholar

• 38

  ChallinorA. J.WatsonJ.LobellD. B.HowdenS. M.SmithD. R.ChhetriN. (2014). A meta-analysis of crop yield under climate change and adaptation.Nat. Clim. Change4287–291. 10.1038/nclimate2153

  • CrossRef
  • Google Scholar

• 39

  ChaudharyK.PoirionO. B.LuL.GarmireL. X. (2018). Deep learning–based multi-omics integration robustly predicts survival in liver cancer.Clin. Cancer Res.241248–1259. 10.1158/1078-0432.CCR-17-0853

  • CrossRef
  • Google Scholar

• 40

  CheZ.PurushothamS.ChoK.SontagD.LiuY. (2018). Recurrent neural networks for multivariate time series with missing values.Sci. Rep.8:6085. 10.1038/s41598-018-24271-9

  • CrossRef
  • Google Scholar

• 41

  ChenD.NeumannK.FriedelS.KilianB.ChenM.AltmannT.et al (2014). Dissecting the phenotypic components of crop plant growth and drought responses based on high-throughput image analysis.Plant Cell264636–4655. 10.1105/tpc.114.129601

  • CrossRef
  • Google Scholar

• 42

  ChenJ.ShiyomiM. (2019). A power law model for analyzing spatial patterns of vegetation abundance in terms of cover, biomass, density, and occurrence: derivation of a common rule.J. Plant Res.132481–497. 10.1007/s10265-019-01116-8

  • CrossRef
  • Google Scholar

• 43

  ChristensenO. F.LundM. S. (2010). Genomic prediction when some animals are not genotyped.Genet. Sel. Evol.42:2. 10.1186/1297-9686-42-2

  • CrossRef
  • Google Scholar

• 44

  CIMMYT (2019). CIMMYT: International Maize and Wheat Improvement Center. Available at: https://www.cimmyt.org/ [Accessed October 1, 2019]

  • Google Scholar

• 45

  ColleauJ. J.DucrocqV.BoichardD.LarroqueH. (1999). Approximate Multitrait BLUP Evaluation to Combine Functional Traits Information. Available at: file: C:/Users/Fabiana/OneDrive-purdue.edu/PhD/Reviewpaper/393-784-1-SM.pdf(accessed June 3, 2019).

  • Google Scholar

• 46

  CoppensF.WuytsN.InzéD.DhondtS. (2017). Unlocking the potential of plant phenotyping data through integration and data-driven approaches.Curr. Opin. Syst. Biol.458–63. 10.1016/j.coisb.2017.07.002

  • CrossRef
  • Google Scholar

• 47

  CrainJ.MondalS.RutkoskiJ.SinghR. P.PolandJ. (2018). Combining high-throughput phenotyping and genomic information to increase prediction and selection accuracy in wheat breeding.Plant Genome111–14. 10.3835/plantgenome2017.05.0043

  • CrossRef
  • Google Scholar

• 48

  CrossaJ.Pérez-RodríguezP.CuevasJ.Montesinos-LópezO.JarquínD.de Los CamposG.et al (2017). Genomic selection in plant breeding: methods, models, and perspectives.Trends Plant Sci.22961–975. 10.1016/j.tplants.2017.08.011

  • CrossRef
  • Google Scholar

• 49

  CullisB. R.SmithA. B.CoombesN. E. (2006). On the design of early generation variety trials with correlated data.J. Agric. Biol. Environ. Stat.11381–393. 10.1198/108571106X154443

  • CrossRef
  • Google Scholar

• 50

  DasA.RushtonP. J.RohilaJ. S. (2017). Metabolomic profiling of soybeans (Glycine max L.) reveals the importance of sugar and nitrogen metabolism under drought and heat stress.Plants6199–208. 10.3390/plants6020021

  • CrossRef
  • Google Scholar

• 51

  DasK.LiJ.WangZ.TongC.FuG.LiY.et al (2011). A dynamic model for genome-wide association studies.Hum. Genet.129629–639. 10.1007/s00439-011-0960-6

  • CrossRef
  • Google Scholar

• 52

  de Almeida FilhoJ. E.GuimarãesJ. F. R.Fonsceca e SilvaF.Vilela de ResendeM. D.MuñozP.KirstM.et al (2019). Genomic prediction of additive and non-additive effects using genetic markers and pedigrees.G392739–2748. 10.1534/g3.119.201004

  • CrossRef
  • Google Scholar

• 53

  de BoorC. (1980). A practical guide to splines.Math. Comput.34:325. 10.2307/2006241

  • CrossRef
  • Google Scholar

• 54

  DiersB. W.SpechtJ.RaineyK. M.CreganP.SongQ.RamasubramanianV.et al (2018). Genetic architecture of soybean yield and agronomic traits.G3 (Bethesda)83367–3375. 10.1534/g3.118.200332

  • CrossRef
  • Google Scholar

• 55

  Durón-BenítezA.-A.WellerJ. I.EzraE. (2018). Using geometric morphometrics for the genetics analysis of shape and size of lactation curves in Israeli first-parity Holstein cattle.J. Dairy Sci.10111132–11142. 10.3168/jds.2018-15209

  • CrossRef
  • Google Scholar

• 56

  DuvickD. N. (2005). Genetic progress in yield of United States maize (Zea mays L.).Maydica50193–202.

  • Google Scholar

• 57

  EastE. M. (1911). The genotype hypothesis and hybridization.Am. Nat.45160–174. 10.1086/279203

  • CrossRef
  • Google Scholar

• 58

  EilersP. H. C.MarxB. D. (1996). Flexible Smoothing with B-splines and Penalties. Available at: https://projecteuclid.org/download/pdf_1/euclid.ss/1038425655(accessed May 24, 2019).

  • Google Scholar

• 59

  EnglishbyT. M.BanosG.MooreK. L.CoffeyM. P.EvansR. D.BerryD. P. (2016). Genetic analysis of carcass traits in beef cattle using random regression models.J. Anim. Sci.941354–1364. 10.2527/jas.2015-0246

  • CrossRef
  • Google Scholar

• 60

  FahlgrenN.GehanM. A.BaxterI. (2015). Lights, camera, action: high-throughput plant phenotyping is ready for a close-up.Curr. Opin. Plant Biol.2493–99. 10.1016/J.PBI.2015.02.006

  • CrossRef
  • Google Scholar

• 61

  FalconerD. S.MackayT. F. C. (1996). Introduction to Quantitative Genetics, 4th Edn. Burnt Mill: Longman.

  • Google Scholar

• 62

  FanR.AlbertP. S.SchistermanE. F. (2012). A discussion of gene-gene and gene-environment interactions and longitudinal genetic analysis of complex traits.Stat. Med.312565–2568. 10.1002/sim.5495

  • CrossRef
  • Google Scholar

• 63

  FeldmanM. J.PaulR. E.BananD.BarrettJ. F.SebastianJ.YeeM. C.et al (2017). Time dependent genetic analysis links field and controlled environment phenotypes in the model C4grass Setaria.PLoS Genet.13:e1006841. 10.1371/journal.pgen.1006841

  • CrossRef
  • Google Scholar

• 64

  FernandezG. C. J. (2019). Repeated measure analysis of line-source sprinkler experiments.HortScience26339–342. 10.21273/hortsci.26.4.339

  • CrossRef
  • Google Scholar

• 65

  FioraniF.RascherU.JahnkeS.SchurrU. (2012). Imaging plants dynamics in heterogenic environments.Curr. Opin. Biotechnol.23227–235. 10.1016/j.copbio.2011.12.010

  • CrossRef
  • Google Scholar

• 66

  FioraniF.SchurrU. (2013). Future scenarios for plant phenotyping.Annu. Rev. Plant Biol.64267–291. 10.1146/annurev-arplant-050312-120137

  • CrossRef
  • Google Scholar

• 67

  FischerR. A.EdmeadesG. O. (2010). Breeding and cereal yield progress.Crop Sci.50S-85–S-98. 10.2135/cropsci2009.10.0564

  • CrossRef
  • Google Scholar

• 68

  FoleyJ. A.RamankuttyN.BraumanK. A.CassidyE. S.GerberJ. S.JohnstonM.et al (2011). Solutions for a cultivated planet.Nature478337–342. 10.1038/nature10452

  • CrossRef
  • Google Scholar

• 69

  FosterJ. J.BarkusE.YavorskyC. (2006). Understanding and using advanced statistics.Choice Rev.4343-5938–43-5938. 10.5860/choice.43-5938

  • CrossRef
  • Google Scholar

• 70

  FurbankR. T.TesterM. (2011). Phenomics - technologies to relieve the phenotyping bottleneck.Trends Plant Sci.16635–644. 10.1016/j.tplants.2011.09.005

  • CrossRef
  • Google Scholar

• 71

  GaoA. W.SchlüterS.BlaserS. R. G. A.ShenJ.VetterleinD. (2019). A shape-based method for automatic and rapid segmentation of roots in soil from X-ray computed tomography images: rootine.Plant Soil4411–13. 10.1007/s11104-019-04053-6

  • CrossRef
  • Google Scholar

• 72

  GaoL.PanH.LiuF.XieX.ZhangZ.HanJ. (2018). “Brain disease diagnosis using deep learning features from longitudinal MR images,” in Lecture Notes in Computer Science (Including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), edsGoebelR.TanakaY.WahlsterW., (Cham: Springer), 327–339. 10.1007/978-3-319-96890-2_27

  • CrossRef
  • Google Scholar

• 73

  GebeyehouG.KnottD. R.BakerR. J. (1982). Rate and duration of grain filling in durum wheat cultivars1.Crop Sci.22:337. 10.2135/cropsci1982.0011183X002200020033x

  • CrossRef
  • Google Scholar

• 74

  GillihamM.AbleJ. A.RoyS. J. (2017). Translating knowledge about abiotic stress tolerance to breeding programmes.Plant J.90898–917. 10.1111/tpj.13456

  • CrossRef
  • Google Scholar

• 75

  GodfrayH. C. J.BeddingtonJ. R.CruteI. R.HaddadL.LawrenceD.MuirJ. F.et al (2010). Food security: the challenge of feeding 9 billion people.Science (80-)327812–818. 10.1126/science.1185383

  • CrossRef
  • Google Scholar

• 76

  GompertzB. (1815). On the nature of the function expressive of the law of human mortality, and on a new mode of determining the value of life contingencies.Proc. R. Soc. Lond.2252–253. 10.1098/rspl.1815.0271

  • CrossRef
  • Google Scholar

• 77

  GottschalkP. G.DunnJ. R. (2005). The five-parameter logistic: a characterization and comparison with the four-parameter logistic.Anal. Biochem.34354–65. 10.1016/j.ab.2005.04.035

  • CrossRef
  • Google Scholar

• 78

  GranierC.VileD. (2014). Phenotyping and beyond: modelling the relationships between traits.Curr. Opin. Plant Biol.1896–102. 10.1016/j.pbi.2014.02.009

  • CrossRef
  • Google Scholar

• 79

  GrosuH.SchaefferL.OltenacuP. A.NormanD.PowellR. L.KremerV.et al (2013). History of Genetic Evaluation Methods in Dairy Cattle. Available at: https://books.google.com/books/about/History_of_Genetic_Evaluation_Methods_in.html?id=dNbsoQEACAAJ(accessed June 6, 2019).

  • Google Scholar

• 80

  GroßkinskyD. K.Jaya SyaifullahS.RoitschT.LimP.SyaifullahS. J.RoitschT. (2018). Integration of multi-omics techniques and physiological phenotyping within a holistic phenomics approach to study senescence in model and crop plants.J. Exp. Bot.69825–844. 10.1093/jxb/erx333

  • CrossRef
  • Google Scholar

• 81

  GuariniA. R.LourencoD. A. L.BritoL. F.SargolzaeiM.BaesC. F.MigliorF.et al (2019a). Genetics and genomics of reproductive disorders in canadian holstein cattle.J. Dairy Sci.1021341–1353. 10.3168/jds.2018-15038

  • CrossRef
  • Google Scholar

• 82

  GuariniA. R.LourencoD. A. L.BritoL. F.SargolzaeiM.BaesC. F.MigliorF.et al (2019b). Use of a single-step approach for integrating foreign information into national genomic evaluation in Holstein cattle.J. Dairy Sci.1028175–8183. 10.3168/jds.2018-15819

  • CrossRef
  • Google Scholar

• 83

  GuoG.ZhaoF.WangY.ZhangY.DuL.SuG. (2014). Comparison of single-trait and multiple-trait genomic prediction models.BMC Genet.15:30. 10.1186/1471-2156-15-30

  • CrossRef
  • Google Scholar

• 84

  GuoT.YangN.TongH.PanQ.YangX.TangJ.et al (2014). Genetic basis of grain yield heterosis in an “immortalized F2” maize population.Theor. Appl. Genet.1272149–2158. 10.1007/s00122-014-2368-x

  • CrossRef
  • Google Scholar

• 85

  HabierD.FernandoR. L.DekkersJ. C. M. (2007). The impact of genetic relationship information on genome-assisted breeding values.Genetics1772389–2397. 10.1534/genetics.107.081190

  • CrossRef
  • Google Scholar

• 86

  HabierD.FernandoR. L.DekkersJ. C. M. (2009). Genomic selection using low-density marker panels.Genetics182343–353. 10.1534/genetics.108.100289

  • CrossRef
  • Google Scholar

• 87

  HarfoucheA. L.JacobsonD. A.KainerD.RomeroJ. C.HarfoucheA. H.Scarascia MugnozzaG.et al (2019). Accelerating climate resilient plant breeding by applying next-generation artificial intelligence.Trends Biotechnol.371217–1235. 10.1016/j.tibtech.2019.05.007

  • CrossRef
  • Google Scholar

• 88

  HearstA. A. (2019). Remote Sensing of Soybean Canopy Cover, Color, and Visible Indicators of Moisture Stress Using Imagery from Unmanned Aircraft Systems, PhD., West Lafayette: Agricultural and Biological Engineering.

  • Google Scholar

• 89

  HendersonC. R. (1974). General flexibility of linear model techniques for sire evaluation.J. Dairy Sci.57963–972. 10.3168/jds.s0022-0302(74)84993-3

  • CrossRef
  • Google Scholar

• 90

  HendersonC. R.QuaasR. L. (1976). Multiple trait evaluation using relatives’, recordsJ. Anim. Sci.431188–1197. 10.2527/jas1976.4361188x

  • CrossRef
  • Google Scholar

• 91

  HenryonM.BergP.SørensenA. C. (2014). Invited review: animal-breeding schemes using genomic information need breeding plans designed to maximise long-term genetic gains.Livest. Sci.16638–47. 10.1016/j.livsci.2014.06.016

  • CrossRef
  • Google Scholar

• 92

  HowardR.JarquinD. (2019). Genomic prediction using canopy coverage image and genotypic information in soybean via a hybrid model.Evol. Bioinform.15:117693431984002. 10.1177/1176934319840026

  • CrossRef
  • Google Scholar

• 93

  HughesA. P.FreemanP. R. (1967). Growth analysis using frequent small harvests.J. Appl. Ecol.4:553. 10.2307/2401356

  • CrossRef
  • Google Scholar

• 94

  HundA.TrachselS.StampP. (2009). Growth of axile and lateral roots of maize: I development of a phenotying platform.Plant Soil325335–349. 10.1007/s11104-009-9984-2

  • CrossRef
  • Google Scholar

• 95

  HurtadoP. X.SchnabelS. K.ZabanA.VeteläinenM.VirtanenE.EilersP. H. C.et al (2012). Dynamics of senescence-related QTLs in potato.Euphytica183289–302. 10.1007/s10681-011-0464-4

  • CrossRef
  • Google Scholar

• 96

  HuynhH.FeldtL. S. (1970). Conditions under which mean square ratios in repeated measurements designs have exact F-distributions.J. Am. Stat. Assoc.651582–1589. 10.1080/01621459.1970.10481187

  • CrossRef
  • Google Scholar

• 97

  JamrozikJ.SchaefferL. R. (1997). Estimates of genetic parameters for a test day model with random regressions for yield traits of first lactation holsteins.J. Dairy Sci.80762–770. 10.3168/jds.S0022-0302(97)75996-4

  • CrossRef
  • Google Scholar

• 98

  JarquinD.HowardR.XavierA.Das ChoudhuryS. (2018). Increasing predictive ability by modeling interactions between environments, genotype and canopy coverage image data for soybeans.Agronomy8:51. 10.3390/agronomy8040051

  • CrossRef
  • Google Scholar

• 99

  JiaY.JanninkJ.-L. L.GoddardM. E. (2012). Multiple-trait genomic selection methods increase genetic value prediction accuracy.Genetics1921513–1522. 10.1534/genetics.112.144246

  • CrossRef
  • Google Scholar

• 100

  JiangJ.XingF.WangC.ZengX.ZouQ. (2019). Investigation and development of maize fused network analysis with multi-omics.Plant Physiol. Biochem.141380–387. 10.1016/j.plaphy.2019.06.016

  • CrossRef
  • Google Scholar

• 101

  JiangJ.ZhangQ.MaL.LiJ.WangZ.LiuJ. F. (2015). Joint prediction of multiple quantitative traits using a Bayesian multivariate antedependence model.Heredity (Edinb)11529–36. 10.1038/hdy.2015.9

  • CrossRef
  • Google Scholar

• 102

  JohannsenW. (1909). Elemente der Exakten Erblichkeitslehre. [Elements of the Exact Theory of Inheritance].Jena: G. Fischer, 10.5962/bhl.title.94247

  • CrossRef
  • Google Scholar

• 103

  JohannsenW. (1911). The genotype conception of heredity.Am. Nat.45129–159. 10.1086/279202

  • CrossRef
  • Google Scholar

• 104

  JonesD. B.PetersonM. L.GengS. (1979). Association between grain filling rate and duration and yield components in rice.Crop Sci.19:641. 10.2135/cropsci1979.0011183x001900050023x

  • CrossRef
  • Google Scholar

• 105

  JulianaP.Montesinos-LópezO. A.CrossaJ.MondalS.González PérezL.PolandJ.et al (2018). Integrating genomic-enabled prediction and high-throughput phenotyping in breeding for climate-resilient bread wheat.Theor. Appl. Genet.132177–194. 10.1007/s00122-018-3206-3

  • CrossRef
  • Google Scholar

• 106

  KangH.NingC.ZhouL.ZhangS.YanQ.LiuJ. F. (2018). Short communication: single-step genomic evaluation of milk production traits using multiple-trait random regression model in Chinese Holsteins.J. Dairy Sci.10111143–11149. 10.3168/jds.2018-15090

  • CrossRef
  • Google Scholar

• 107

  KangH.ZhouL.MrodeR.ZhangQ.LiuJ. F. (2017). Incorporating the single-step strategy into a random regression model to enhance genomic prediction of longitudinal traits.Heredity (Edinb)119459–467. 10.1038/hdy.2016.91

  • CrossRef
  • Google Scholar

• 108

  KirkpatrickM.HeckmanN. (1989). A quantitative genetic model for growth, shape, reaction norms, and other infinite-dimensional characters.J. Math. Biol.27429–450. 10.1007/BF00290638

  • CrossRef
  • Google Scholar

• 109

  KirkpatrickM.LofsvoldD.BulmerM. (1990). Analysis of the inheritance, selection and evolution of growth trajectories.Genetics124979–993.

  • Google Scholar

• 110

  KnochD.AbbadiA.GrandkeF.MeyerR. C.SamansB.WernerC. R.et al (2020). Strong temporal dynamics of QTL action on plant growth progression revealed through high-throughput phenotyping in canola.Plant Biotechnol. J.1868–82. 10.1111/pbi.13171

  • CrossRef
  • Google Scholar

• 111

  KoivulaM.StrandénI.PösöJ.AamandG. P.MäntysaariE. A. (2015). Single-step genomic evaluation using multitrait random regression model and test-day data.J. Dairy Sci.982775–2784. 10.3168/jds.2014-8975

  • CrossRef
  • Google Scholar

• 112

  KonietschkeF.BathkeA. C.HarrarS. W.PaulyM. (2015). Parametric and nonparametric bootstrap methods for general MANOVA.J. Multivar. Anal.140291–301. 10.1016/j.jmva.2015.05.001

  • CrossRef
  • Google Scholar

• 113

  KoutroubasS. D.PapakostaD. K. (2010). Seed filling patterns of safflower: genotypic and seasonal variations and association with other agronomic traits.Ind. Crops Prod.3171–76. 10.1016/J.INDCROP.2009.09.014

  • CrossRef
  • Google Scholar

• 114

  KrishnamoorthyK.LuF. (2010). A parametric bootstrap solution to the MANOVA under heteroscedasticity.J. Stat. Comput. Simul.80873–887. 10.1080/00949650902822564

  • CrossRef
  • Google Scholar

• 115

  KrishnamoorthyK.YuJ. (2012). Multivariate Behrens-Fisher problem with missing data.J. Multivar. Anal.105141–150. 10.1016/j.jmva.2011.08.019

  • CrossRef
  • Google Scholar

• 116

  KwakI.-Y.MooreC. R.SpaldingE. P.BromanK. W. (2016). Mapping quantitative trait loci underlying function-valued traits using functional principal component analysis and multi-trait mapping.G3679–86. 10.1534/G3.115.024133

  • CrossRef
  • Google Scholar

• 117

  KwakI.-Y. Y.MooreC. R.SpaldingE. P.BromanK. W. (2014). A simple regression-based method to map quantitative trait loci underlying function-valued phenotypes.Genetics1971409–1416. 10.1534/genetics.114.166306

  • CrossRef
  • Google Scholar

• 118

  LangeC. E.FederizziL. C. (2009). Estimation of soybean genetic progress in the South of Brazil using multi-environmental yield trials.Sci. Agric.66309–316. 10.1590/s0103-90162009000300005

  • CrossRef
  • Google Scholar

• 119

  LangridgeP.FleuryD. (2011). Making the most of “omics” for crop breeding.Trends Biotechnol.2933–40. 10.1016/j.tibtech.2010.09.006

  • CrossRef
  • Google Scholar

• 120

  LecunY.BengioY.HintonG. (2015). Deep learning.Nature521436–444. 10.1038/nature14539

  • CrossRef
  • Google Scholar

• 121

  LeeG.NhoK.KangB.SohnK. A.KimD.WeinerM. W.et al (2019). Predicting Alzheimer’s disease progression using multi-modal deep learning approach.Sci. Rep.9:1952. 10.1038/s41598-018-37769-z

  • CrossRef
  • Google Scholar

• 122

  LegarraA.AguilarI.MisztalI. (2009). A relationship matrix including full pedigree and genomic information.J. Dairy Sci.924656–4663. 10.3168/jds.2009-2061

  • CrossRef
  • Google Scholar

• 123

  LegarraA.ChristensenO. F.AguilarI.MisztalI. (2014). Single step, a general approach for genomic selection.Livest. Sci.16654–65. 10.1016/j.livsci.2014.04.029

  • CrossRef
  • Google Scholar

• 124

  LeonJ.GeislerG. (1994). Variation in rate and duration of growth among spring barley cultivars1.Plant Breed.112199–208. 10.1111/j.1439-0523.1994.tb00671.x

  • CrossRef
  • Google Scholar

• 125

  LiD.HuangZ.SongS.XinY.MaoD.LvQ.et al (2016). Integrated analysis of phenome, genome, and transcriptome of hybrid rice uncovered multiple heterosis-related loci for yield increase.Proc. Natl. Acad. Sci. U.S.A.113E6026–E6035. 10.1073/pnas.1610115113

  • CrossRef
  • Google Scholar

• 126

  LiL.ZhangQ.HuangD. (2014). A review of imaging techniques for plant phenotyping.Sensors (Switzerland)1420078–20111. 10.3390/s141120078

  • CrossRef
  • Google Scholar

• 127

  LiZ.SillanpääM. J. (2013). A bayesian nonparametric approach for mapping dynamic quantitative traits.Genetics194997–1016. 10.1534/genetics.113.152736

  • CrossRef
  • Google Scholar

• 128

  LiZ.SillanpääM. J. (2015). Dynamic quantitative trait locus analysis of plant phenomic data.20822–833. 10.1016/j.tplants.2015.08.012

  • CrossRef
  • Google Scholar

• 129

  LillehammerM.ØdegårdJ.MeuwissenT. H. (2007). Random regression models for detection of gene by environment interaction.Genet. Sel. Evol.39:105. 10.1186/1297-9686-39-2-105

  • CrossRef
  • Google Scholar

• 130

  LittellR. C. (1990). “Analysis of repeated measures data,” in Conference on Applied Statistic in Agriculture, Gainesville, FL, 10.1007/978-981-10-3794-8

  • CrossRef
  • Google Scholar

• 131

  LittellR. C.HenryP. R.AmmermanC. B. (1998). Statistical analysis of repeated measures data using SAS procedures.J. Anim. Sci.761216-1231. 10.2527/1998.7641216x

  • CrossRef
  • Google Scholar

• 132

  LittellR. C.PendergastJ.NatarajanR. (2000). Tutorial in Biostatics: modelling covariance structure in the analysis of repeated measures data.Stat. Med.191793–1819. 10.1002/1097-0258(20000715)19:13<1793::aid-sim482>3.0.co;2-q

  • CrossRef
  • Google Scholar

• 133

  LiuX.DongX.XueQ.LeskovarD. I.JifonJ.ButnorJ. R.et al (2018a). Ground penetrating radar (GPR) detects fine roots of agricultural crops in the field.Plant Soil423517–531. 10.1007/s11104-017-3531-3

  • CrossRef
  • Google Scholar

• 134

  LiuX.WangH.WangH.GuoZ.XuX.LiuJ.et al (2018b). Factors affecting genomic selection revealed by empirical evidence in maize.Crop J.6341–352. 10.1016/j.cj.2018.03.005

  • CrossRef
  • Google Scholar

• 135

  LiuX.WangH.HuX.LiK.LiuZ.WuY.et al (2019). Improving genomic selection with quantitative trait loci and nonadditive effects revealed by empirical evidence in maize.Front. Plant Sci.10:1129. 10.3389/fpls.2019.01129

  • CrossRef
  • Google Scholar

• 136

  MaC. X.CasellaG.WuR. (2002). Functional mapping of quantitative trait loci underlying the character process: a theoretical framework.Genetics1611751–1762.

  • Google Scholar

• 137

  MaL.ShiY.SiemianowskiO.YuanB.EgnerT. K.MirnezamiS. V.et al (2019). Hydrogel-based transparent soils for root phenotyping in vivo.Proc. Natl. Acad. Sci. U.S.A.16611063–11068. 10.1073/pnas.1820334116

  • CrossRef
  • Google Scholar

• 138

  MaW.QiuZ.SongJ.LiJ.ChengQ.ZhaiJ.et al (2018). A deep convolutional neural network approach for predicting phenotypes from genotypes.Planta2481307–1318. 10.1007/s00425-018-2976-9

  • CrossRef
  • Google Scholar

• 139

  MacciottaN. P. P.BiffaniS.BernabucciU.LaceteraN.VitaliA.Ajmone-MarsanP.et al (2017). Derivation and genome-wide association study of a principal component-based measure of heat tolerance in dairy cattle.J. Dairy Sci.1004683–4697. 10.3168/jds.2016-12249

  • CrossRef
  • Google Scholar

• 140

  MacciottaN. P. P.GaspaG.BombaL.VicarioD.DimauroC.CellesiM.et al (2015). Genome-wide association analysis in Italian simmental cows for lactation curve traits using a low-density (7K) SNP panel.J. Dairy Sci.988175–8185. 10.3168/jds.2015-9500

  • CrossRef
  • Google Scholar

• 141

  MacciottaN. P. P.VicarioD.Cappio-BorlinoA. (2006). Use of multivariate analysis to extract latent variables related to level of production and lactation persistency in dairy cattle.J. Dairy Sci.893188–3194. 10.3168/jds.S0022-0302(06)72593-0

  • CrossRef
  • Google Scholar

• 142

  MarquetP. A.QuiñonesR. A.AbadesS.LabraF.TognelliM.ArimM.et al (2005). Scaling and power-laws in ecological systems.J. Exp. Biol.2081749–1769. 10.1242/jeb.01588

  • CrossRef
  • Google Scholar

• 143

  MeadeK. A.CooperM.BeavisW. D. (2013). Modeling biomass accumulation in maize kernels.F. Crop. Res.15192–100. 10.1016/j.fcr.2013.07.014

  • CrossRef
  • Google Scholar

• 144

  MetznerR.EggertA.van DusschotenD.PflugfelderD.GerthS.SchurrU.et al (2015). Direct comparison of MRI and X-ray CT technologies for 3D imaging of root systems in soil: potential and challenges for root trait quantification.Plant Methods11:17. 10.1186/s13007-015-0060-z

  • CrossRef
  • Google Scholar

• 145

  MeuwissenT. H. (2009). Accuracy of breeding values of “unrelated” individuals predicted by dense SNP genotyping.Genet. Sel. Evol.41:35. 10.1186/1297-9686-41-35

  • CrossRef
  • Google Scholar

• 146

  MeuwissenT. H. E.HayesB. J.GoddardM. E. (2001). Prediction of total genetic value using genome-wide dense marker maps.Genetics1571819–1829.

  • Google Scholar

• 147

  MeuwissenT. H. E.SvendsenM.SolbergT.ØdegårdJ. (2015). Genomic predictions based on animal models using genotype imputation on a national scale in Norwegian Red cattle.Genet. Sel. Evol.47:79. 10.1186/s12711-015-0159-8

  • CrossRef
  • Google Scholar

• 148

  MeyerK. (1998). Estimating covariance functions for longitudinal data using a random regression model.Genet. Sel. Evol.30221–240. 10.1186/1297-9686-30-3-221

  • CrossRef
  • Google Scholar

• 149

  MeyerK. (2005a). Advances in methodology for random regression analyses.Austr. J. Exp. Agric.45847–858. 10.1071/EA05040

  • CrossRef
  • Google Scholar

• 150

  MeyerK. (2005b). Random regression analyses using B-splines to model growth of Australian Angus cattle.Genet. Sel. Evol.37:473. 10.1186/1297-9686-37-6-473

  • CrossRef
  • Google Scholar

• 151

  MeyerK. (2019). “Bending” and beyond: better estimates of quantitative genetic parameters?J. Anim. Breed. Genet.136243–251. 10.1111/jbg.12386

  • CrossRef
  • Google Scholar

• 152

  MeyerK.HillW. G. (1997). Estimation of genetic and phenotypic covariance functions for longitudinal or “repeated” records by restricted maximum likelihood.Livest. Prod. Sci.47185–200. 10.1016/S0301-6226(96)01414-5

  • CrossRef
  • Google Scholar

• 153

  MeyerK.KirkpatrickM. (2005). Up hill, down dale: quantitative genetics of curvaceous traits.Philos. Trans. R. Soc. B Biol. Sci.3601443–1455. 10.1098/rstb.2005.1681

  • CrossRef
  • Google Scholar

• 154

  MigliorF.FlemingA.MalchiodiF.BritoL. F.MartinP.BaesC. F. (2017). A 100-Year review: identification and genetic selection of economically important traits in dairy cattle.J. Dairy Sci.10010251–10271. 10.3168/jds.2017-12968

  • CrossRef
  • Google Scholar

• 155

  MisztalI.LegarraA.AguilarI. (2009). Computing procedures for genetic evaluation including phenotypic, full pedigree, and genomic information.J. Dairy Sci.924648–4655. 10.3168/JDS.2009-2064

  • CrossRef
  • Google Scholar

• 156

  MomenM.CampbellM. T.WaliaH.MorotaG. (2019). Predicting longitudinal traits derived from high-throughput phenomics in contrasting environments using genomic legendre polynomials and B-splines.G3 (Bethesda)93369–3380. 10.1534/g3.119.400346

  • CrossRef
  • Google Scholar

• 157

  MonirM. M.ZhuJ. (2018). Dominance and epistasis interactions revealed as important variants for leaf traits of maize NAM population.Front. Plant Sci.9:627. 10.3389/fpls.2018.00627

  • CrossRef
  • Google Scholar

• 158

  MontesJ. M.MelchingerA. E.ReifJ. C. (2007). Novel throughput phenotyping platforms in plant genetic studies.Trends Plant Sci.12433–436. 10.1016/j.tplants.2007.08.006

  • CrossRef
  • Google Scholar

• 159

  Montesinos-LópezA.Montesinos-LópezO. A.GianolaD.CrossaJ.Hernández-SuárezC. M. (2018). Multi-environment genomic prediction of plant traits using deep learners with dense architecture.G383813–3828. 10.1534/g3.118.200740

  • CrossRef
  • Google Scholar

• 160

  Montesinos-LópezO. A.Montesinos-LópezA.CrossaJ.GianolaD.Hernández-SuárezC. M.Martín-VallejoJ. (2018). Multi-trait, multi-environment deep learning modeling for genomic-enabled prediction of plant traits.G383829–3840. 10.1534/g3.118.200728

  • CrossRef
  • Google Scholar

• 161

  Montesinos-LópezO. A.Martín-VallejoJ.CrossaJ.GianolaD.Hernández-SuárezC. M.Montesinos-LópezA.et al (2019). New deep learning genomic-based prediction model for multiple traits with binary, ordinal, and continuous phenotypes.G3 (Bethesda)91545–1556. 10.1534/g3.119.300585

  • CrossRef
  • Google Scholar

• 162

  Montesinos-LópezO. A.Montesinos-LópezA.CrossaJ.los CamposG.AlvaradoG.SuchismitaM.et al (2017). Predicting grain yield using canopy hyperspectral reflectance in wheat breeding data.Plant Methods13:4. 10.1186/s13007-016-0154-2

  • CrossRef
  • Google Scholar

• 163

  Montesinos-LópezO. A.Montesinos-LópezA.CrossaJ.ToledoF. H.Pérez-HernándezO.EskridgeK. M.et al (2016). A genomic bayesian multi-trait and multi-environment model.G3 (Bethesda)62725–2744. 10.1534/g3.116.032359

  • CrossRef
  • Google Scholar

• 164

  MoreiraF. F.HearstA. A.CherkauerK. A.RaineyK. M. (2019). Improving the efficiency of soybean breeding with high-throughput canopy phenotyping.Plant Methods15:139. 10.1186/s13007-019-0519-4

  • CrossRef
  • Google Scholar

• 165

  MrodeR. (2014). Linear Models for the Prediction of Animal Breeding Values, ed.MrodeR.Wallingford: CABI, 10.1079/9781780643915.0000

  • CrossRef
  • Google Scholar

• 166

  MuirB. L.KistemakerG.JamrozikJ.CanavesiF. (2007). Genetic parameters for a multiple-trait multiple-lactation random regression test-day model in italian holsteins.J. Dairy Sci.901564–1574. 10.3168/jds.S0022-0302(07)71642-9

  • CrossRef
  • Google Scholar

• 167

  MurayaM. M.ChuJ.ZhaoY.JunkerA.KlukasC.ReifJ. C.et al (2017). Genetic variation of growth dynamics in maize (Zea mays L.) revealed through automated non-invasive phenotyping.Plant J.89366–380. 10.1111/tpj.13390

  • CrossRef
  • Google Scholar

• 168

  NeathA. A.CavanaughJ. E. (2012). The bayesian information criterion: background, derivation, and applications.Wiley Interdiscip. Rev. Comput. Stat.4199–203. 10.1002/wics.199

  • CrossRef
  • Google Scholar

• 169

  NeilsonE. H.EdwardsA. M.BlomstedtC. K.BergerB.MøllerB. L.GleadowR. M. (2015). Utilization of a high-throughput shoot imaging system to examine the dynamic phenotypic responses of a C4 cereal crop plant to nitrogen and water deficiency over time.J. Exp. Bot.661817–1832. 10.1093/jxb/eru526

  • CrossRef
  • Google Scholar

• 170

  NeumannK.ZhaoY.ChuJ.KeilwagenJ.ReifJ. C.KilianB.et al (2017). Genetic architecture and temporal patterns of biomass accumulation in spring barley revealed by image analysis.BMC Plant Biol.17:137. 10.1186/s12870-017-1085-4

  • CrossRef
  • Google Scholar

• 171

  NingC.KangH.ZhouL.WangD.WangH.WangA.et al (2017). Performance gains in genome-wide association studies for longitudinal traits via modeling time-varied effects.Sci. Rep.7:590. 10.1038/s41598-017-00638-2

  • CrossRef
  • Google Scholar

• 172

  NingC.WangD.ZhengX.ZhangQ.ZhangS.MrodeR.et al (2018). Eigen decomposition expedites longitudinal genome-wide association studies for milk production traits in Chinese Holstein.Genet. Sel. Evol.50:12. 10.1186/s12711-018-0383-0

  • CrossRef
  • Google Scholar

• 173

  NobreP. R. C.MisztalI.TsurutaS.BertrandJ. K.SilvaL. O. C.LopesP. S. (2003). Analyses of growth curves of Nellore cattle by multiple-trait and random regression models.J. Anim. Sci.81918–926. 10.2527/2003.814918x

  • CrossRef
  • Google Scholar

• 174

  NolanE.SantosP. (2012). The contribution of genetic modification to changes in corn yield in the United States.Am. J. Agric. Econ.941171–1188. 10.1093/ajae/aas069

  • CrossRef
  • Google Scholar

• 175

  OliveiraH. R.BritoL. F.LourencoD. A. L.SilvaF. F.JamrozikJ.SchaefferL. R.et al (2019a). Invited review: advances and applications of random regression models: from quantitative genetics to genomics.J. Dairy Sci.1027664–7683. 10.3168/jds.2019-16265

  • CrossRef
  • Google Scholar

• 176

  OliveiraH. R.BritoL. F.SilvaF. F.LourencoD. A. L.JamrozikJ.SchenkelF. S. (2019b). Genomic prediction of lactation curves for milk, fat, protein, and somatic cell score in Holstein cattle.J. Dairy Sci.102452–463. 10.3168/jds.2018-15159

  • CrossRef
  • Google Scholar

• 177

  OliveiraH. R.e SilvaF. F.Barbosa da SilvaM. V. G.Barbosa Dias de SiqueiraO. H. G.MachadoM. A.do Carmo PanettoJ. C. (2017). Bayesian models combining legendre and b-spline polynomials for genetic analysis of multiple lactations in Gyr cattle.Livest. Sci.20178–84. 10.1016/j.livsci.2017.05.007

  • CrossRef
  • Google Scholar

• 178

  OliveiraH. R.SilvaF. F.BritoL. F.GuariniA. R.JamrozikJ.SchenkelF. S. (2018). Comparing deregression methods for genomic prediction of test-day traits in dairy cattle.J. Anim. Breed. Genet.13597–106. 10.1111/jbg.12317

  • CrossRef
  • Google Scholar

• 179

  OliveiraH. R. R.LourencoD. A. L. A. L.MasudaY.MisztalI.TsurutaS.JamrozikJ.et al (2019c). Application of single-step genomic evaluation using multiple-trait random regression test-day models in dairy cattle.J. Dairy Sci.1022365–2377. 10.3168/jds.2018-15466

  • CrossRef
  • Google Scholar

• 180

  OliveiraH. R.SilvaF. F.SiqueiraO. H.SouzaN. O.JunqueiraV. S.ResendeM. D. V. V.et al (2016). Combining different functions to describe milk, fat, and protein yield in goats using Bayesian multiple-trait random regression models.J. Anim. Sci.941865–1874. 10.2527/jas.2015-0150

  • CrossRef
  • Google Scholar

• 181

  PaineC. E. T. T.MarthewsT. R.VogtD. R.PurvesD.ReesM.HectorA.et al (2012). How to fit nonlinear plant growth models and calculate growth rates: an update for ecologists.Methods Ecol. Evol.3245–256. 10.1111/j.2041-210X.2011.00155.x

  • CrossRef
  • Google Scholar

• 182

  PandeyV.SinghM.PandeyD.KumarA. (2018). Integrated proteomics, genomics, metabolomics approaches reveal oxalic acid as pathogenicity factor in Tilletia indica inciting Karnal bunt disease of wheat.Sci. Rep.8:7826. 10.1038/s41598-018-26257-z

  • CrossRef
  • Google Scholar

• 183

  PattersonJ.GibsonA. (2017). Deep Learning: A Practitioner’s Approach.Sebastopol, CA: O’Reilly Media.

  • Google Scholar

• 184

  PauliD.Andrade-SanchezP.Carmo-SilvaA. E.GazaveE.FrenchA. N.HeunJ.et al (2016a). Field-based high-throughput plant phenotyping reveals the temporal patterns of quantitative trait loci associated with stress-responsive traits in cotton.G36:865. 10.1534/G3.115.023515

  • CrossRef
  • Google Scholar

• 185

  PauliD.ChapmanS. C.BartR.ToppC. N.Lawrence-DillC. J.PolandJ.et al (2016b). The quest for understanding phenotypic variation via integrated approaches in the field environment.Plant Physiol.17200592.2016. 10.1104/pp.16.00592

  • CrossRef
  • Google Scholar

• 186

  PembletonL. W.DraytonM. C.BainM.BaillieR. C.InchC.SpangenbergG. C.et al (2016). Targeted genotyping-by-sequencing permits cost-effective identification and discrimination of pasture grass species and cultivars.Theor. Appl. Genet.129991–1005. 10.1007/s00122-016-2678-2

  • CrossRef
  • Google Scholar

• 187

  PereiraR. J.BignardiA. B.El FaroL.VernequeR. S.Vercesi FilhoA. E.AlbuquerqueL. G. (2013). Random regression models using legendre polynomials or linear splines for test-day milk yield of dairy Gyr (Bos indicus) cattle.J. Dairy Sci.96565–574. 10.3168/JDS.2011-5051

  • CrossRef
  • Google Scholar

• 188

  Pérez-BuenoM. L.PinedaM.CabezaF. M.BarónM. (2016). Multicolor fluorescence imaging as a candidate for disease detection in plant phenotyping.Front. Plant Sci.7:1790. 10.3389/fpls.2016.01790

  • CrossRef
  • Google Scholar

• 189

  Pérez-EncisoM.ZingarettiL. M. (2019). A guide for using deep learning for complex trait genomic prediction.Genes (Basel)10:E553. 10.3390/genes10070553

  • CrossRef
  • Google Scholar

• 190

  PfeiferJ.KirchgessnerN.ColombiT.WalterA. (2015). Rapid phenotyping of crop root systems in undisturbed field soils using X-ray computed tomography.Plant Methods11:41. 10.1186/s13007-015-0084-4

  • CrossRef
  • Google Scholar

• 191

  PflugfelderD.MetznerR.DusschotenD.ReichelR.JahnkeS.KollerR. (2017). Non-invasive imaging of plant roots in different soils using magnetic resonance imaging (MRI).Plant Methods13:102. 10.1186/s13007-017-0252-9

  • CrossRef
  • Google Scholar

• 192

  PiephoH. P.BüchseA. B.RichterC.BüchseD. A. B.BüchseA.RichterC. (2004). A mixed modelling aproach for randomized experiments with repeated measures.J. Agron. Crop Sci.190230–247. 10.1111/j.1439-037X.2004.00097.x

  • CrossRef
  • Google Scholar

• 193

  PiephoH. P.MöhringJ.MelchingerA. E.BüchseA. (2008). BLUP for phenotypic selection in plant breeding and variety testing.Euphytica161209–228. 10.1007/s10681-007-9449-8

  • CrossRef
  • Google Scholar

• 194

  PinheiroJ.BatesD. (2000). Mixed-Effects Models in S and S-PLUS.New York, NY: Springer-Verlag, 10.1007/b98882

  • CrossRef
  • Google Scholar

• 195

  PoorterH. (1989). Plant growth analysis: towards a synthesis of the classical and the functional approach.Physiol. Plant.75237–244. 10.1111/j.1399-3054.1989.tb06175.x

  • CrossRef
  • Google Scholar

• 196

  PoorterH.AntenN. P. R.MarcelisL. F. M. (2013). Physiological mechanisms in plant growth models: do we need a supra-cellular systems biology approach?Plant Cell Environ.361673–1690. 10.1111/pce.12123

  • CrossRef
  • Google Scholar

• 197

  PromislowD. E. L.TatarM.KhazaeliA. A.CurtsingerJ. W. (1996). Age-specific patterns of genetic variance in Drosophila melanogaster. I. mortality.Genetics143839–848.

  • Google Scholar

• 198

  RahamanM. M.ChenD.GillaniZ.KlukasC.ChenM. (2015). Advanced phenotyping and phenotype data analysis for the study of plant growth and development.Front. Plant Sci.6:619. 10.3389/fpls.2015.00619

  • CrossRef
  • Google Scholar

• 199

  RayD. K.MuellerN. D.WestP. C.FoleyJ. A. (2013). Yield trends are insufficient to double global crop production by 2050.PLoS ONE8:e66428. 10.1371/journal.pone.0066428

  • CrossRef
  • Google Scholar

• 200

  ReynoldsM.ChapmanS.Crespo-HerreraL.MoleroG.MondalS.PequenoD. N. L.et al (2020). Breeder friendly phenotyping.Plant Sci.10.1016/j.plantsci.2019.110396

  • CrossRef
  • Google Scholar

• 201

  RichardsF. J. (1959). A flexible growth function for empirical use.J. Exp. Bot.10290–301. 10.1093/jxb/10.2.290

  • CrossRef
  • Google Scholar

• 202

  RogersJ.ChenP.ShiA.ZhangB.ScabooA.SmithS. F.et al (2015). Agronomic performance and genetic progress of selected historical soybean varieties in the southern USA.Plant Breed.13485–93. 10.1111/pbr.12222

  • CrossRef
  • Google Scholar

• 203

  RowellJ. G.WaltersD. E. (1976). Analysing data with repeated observations on each experimental unit.J. Agric. Sci.87423–432. 10.1017/S0021859600027763

  • CrossRef
  • Google Scholar

• 204

  RutkoskiJ.PolandJ.MondalS.AutriqueE.PérezL. G.CrossaJ.et al (2016). Canopy temperature and vegetation indices from high-throughput phenotyping improve accuracy of pedigree and genomic selection for grain yield in wheat.G362799–2808. 10.1534/g3.116.032888

  • CrossRef
  • Google Scholar

• 205

  SaganV.MaimaitijiangM.SidikeP.EblimitK.PetersonK. T.HartlingS.et al (2019). UAV-based high resolution thermal imaging for vegetation monitoring, and plant phenotyping using ICI 8640 P, FLIR Vue Pro R 640, and thermomap cameras.Remote Sens.11:330. 10.3390/rs11030330

  • CrossRef
  • Google Scholar

• 206

  SantanaM. L.BignardiA. B.PereiraR. J.Menéndez-BuxaderaA.El FaroL. (2016). Random regression models to account for the effect of genotype by environment interaction due to heat stress on the milk yield of Holstein cows under tropical conditions.J. Appl. Genet.57119–127. 10.1007/s13353-015-0301-x

  • CrossRef
  • Google Scholar

• 207

  SchaefferL. R. (2004). Application of random regression models in animal breeding.Livest. Prod. Sci.8635–45. 10.1016/S0301-6226(03)00151-9

  • CrossRef
  • Google Scholar

• 208

  SchaefferL. R. (2016). Random Regression Models. Available online at: https://www.luke.fi/wp-content/uploads/2018/01/Random-regression-model.pdf(accessed June 3, 2019).

  • Google Scholar

• 209

  ScheinerS. M.GurevitchJ. (2001). Design and Analysis of Ecological Experiments.Oxford: Oxford University Press.

  • Google Scholar

• 210

  SchowengerdtR. A. (2012). Remote Sensing: Models and Methods for Image Processing.Amsterdam: Elsevier, 10.1016/j.jenvman.2011.10.007

  • CrossRef
  • Google Scholar

• 211

  SchragT. A.SchipprackW.MelchingerA. E. (2019). Across-years prediction of hybrid performance in maize using genomics.Theor. Appl. Genet.132933–946. 10.1007/s00122-018-3249-5

  • CrossRef
  • Google Scholar

• 212

  ShanahanP. W.BinleyA.WhalleyW. R.WattsC. W. (2015). The use of electromagnetic induction to monitor changes in soil moisture profiles beneath different wheat genotypes.Soil Sci. Soc. Am. J.79:459. 10.2136/sssaj2014.09.0360

  • CrossRef
  • Google Scholar

• 213

  ShengJ.ZhengX.WangJ.ZengX.ZhouF.JinS.et al (2017). Transcriptomics and proteomics reveal genetic and biological basis of superior biomass crop Miscanthus.Sci. Rep.7:13777. 10.1038/s41598-017-14151-z

  • CrossRef
  • Google Scholar

• 214

  ShiP.TangL.LinC.LiuL.WangH.CaoW.et al (2015). Modeling the effects of post-anthesis heat stress on rice phenology.F. Crop. Res.17726–36. 10.1016/j.fcr.2015.02.023

  • CrossRef
  • Google Scholar

• 215

  SinghA.GanapathysubramanianB.SinghA. K. A.SarkarS. (2016). Machine learning for high-throughput stress phenotyping in plants.Trends Plant Sci.21110–124. 10.1016/J.TPLANTS.2015.10.015

  • CrossRef
  • Google Scholar

• 216

  SongQ.YanL.QuigleyC.JordanB. D.FickusE.SchroederS.et al (2017). Genetic characterization of the soybean nested association mapping population.Plant Genome101–14. 10.3835/plantgenome2016.10.0109

  • CrossRef
  • Google Scholar

• 217

  SpeidelS. E. (2011). Random Regression Models for the Prediction of Days to Finish in Beef Cattle. Available at: https://mountainscholar.org/bitstream/handle/10217/69294/Speidel_colostate_0053A_10777.pdf?sequence=1(accessed July 8, 2019).

  • Google Scholar

• 218

  SperisenP.CominettiO.MartinF. P. J. (2015). Longitudinal omics modeling and integration in clinical metabonomics research: challenges in childhood metabolic health research.Front. Mol. Biosci.2:44. 10.3389/fmolb.2015.00044

  • CrossRef
  • Google Scholar

• 219

  StinchcombeJ. R.KirkpatrickM. (2012). Genetics and evolution of function-valued traits: understanding environmentally responsive phenotypes.Trends Ecol. Evol.27637–647. 10.1016/j.tree.2012.07.002

  • CrossRef
  • Google Scholar

• 220

  StoskopfN. C.TomesD. T.ChristieB. R. (1994). Plant breeding: theory and practice.Choice Rev.3232-0301–32-0301. 10.5860/choice.32-0301

  • CrossRef
  • Google Scholar

• 221

  SunJ.PolandJ. A.MondalS.CrossaJ.JulianaP.SinghR. P.et al (2019). High-throughput phenotyping platforms enhance genomic selection for wheat grain yield across populations and cycles in early stage.Theor. Appl. Genet.1321705–1720. 10.1007/s00122-019-03309-0

  • CrossRef
  • Google Scholar

• 222

  SunJ.RutkoskiJ. E.PolandJ. A.CrossaJ.JanninkJ.-L.SorrellsM. E. (2017). Multitrait, random regression, or simple repeatability model in high-throughput phenotyping data improve genomic prediction for wheat grain yield.Plant Genome101–15. 10.3835/plantgenome2016.11.0111

  • CrossRef
  • Google Scholar

• 223

  SunS.LiC.PatersonA. H.JiangY.XuR.RobertsonJ. S.et al (2018). In-field high throughput phenotyping and cotton plant growth analysis using LiDAR.Front. Plant Sci.9:16. 10.3389/fpls.2018.00016

  • CrossRef
  • Google Scholar

• 224

  SungJ. M.ChoI.-J.SungD.KimS.KimH. C.ChaeM.-H.et al (2019). Development and verification of prediction models for preventing cardiovascular diseases.PLoS ONE14:e0222809. 10.1371/journal.pone.0222809

  • CrossRef
  • Google Scholar

• 225

  SuravajhalaP.KogelmanL. J. A.KadarmideenH. N. (2016). Multi-omic data integration and analysis using systems genomics approaches: methods and applications In animal production, health and welfare.Genet. Sel. Evol.48:38. 10.1186/s12711-016-0217-x

  • CrossRef
  • Google Scholar

• 226

  SzydaJ.KomisarekJ.AntkowiakI. (2014). Modelling effects of candidate genes on complex traits as variables over time.Anim. Genet.45322–328. 10.1111/age.12144

  • CrossRef
  • Google Scholar

• 227

  TaiA. P. K.MartinM. V.HealdC. L. (2014). Threat to future global food security from climate change and ozone air pollution.Nat. Clim. Change4817–821. 10.1038/nclimate2317

  • CrossRef
  • Google Scholar

• 228

  TardieuF.Cabrera-BosquetL.PridmoreT.BennettM. (2017). Plant phenomics, from sensors to knowledge.Curr. Biol.27R770–R783. 10.1016/j.cub.2017.05.055

  • CrossRef
  • Google Scholar

• 229

  TessmerO. L.JiaoY.CruzJ. A.KramerD. M.ChenJ. (2013). Functional approach to high-throughput plant growth analysis.BMC Syst. Biol.7:S17. 10.1186/1752-0509-7-S6-S17

  • CrossRef
  • Google Scholar

• 230

  TesterM.LangridgeP. (2010). Breeding technologies to increase crop production in a changing world.Science (80-)327818–822. 10.1126/science.1183700

  • CrossRef
  • Google Scholar

• 231

  ThompsonR.MeyerK. (1986). A review of theoretical aspects in the estimation of breeding values for multi-trait selection.Livest. Prod. Sci.15299–313. 10.1016/0301-6226(86)90071-0

  • CrossRef
  • Google Scholar

• 232

  ThornleyJ. H. M.FranceJ.CannellM. G. R. (2005). An open-ended logistic-based growth function.Ecol. Modell.184257–261. 10.1016/j.ecolmodel.2004.10.007

  • CrossRef
  • Google Scholar

• 233

  TjørveK. M. C.TjørveE. (2017). The use of gompertz models in growth analyses, and new gompertz-model approach: an addition to the unified-richards family.PLoS ONE12:e0178691. 10.1371/journal.pone.0178691

  • CrossRef
  • Google Scholar

• 234

  ToenniessenG. H. (2002). Crop genetic improvement for enhanced human nutrition.J. Nutr.1322943S–2946S. 10.1093/jn/132.9.2943s

  • CrossRef
  • Google Scholar

• 235

  ToppC. N.Iyer-PascuzziA. S.AndersonJ. T.LeeC.-R.ZurekP. R.SymonovaO.et al (2013). 3D phenotyping and quantitative trait locus mapping identify core regions of the rice genome controlling root architecture.Proc. Natl. Acad. Sci. U.S.A.110E1695–E1704. 10.1073/pnas.1304354110

  • CrossRef
  • Google Scholar

• 236

  TrabelsiA.ChaabaneM.Ben-HurA. (2019). Comprehensive evaluation of deep learning architectures for prediction of DNA/RNA sequence binding specificities.Bioinformatics (Narnia)35i269–i277. 10.1093/bioinformatics/btz339

  • CrossRef
  • Google Scholar

• 237

  TrachselS.KaepplerS. M.BrownK. M.LynchJ. P. (2011). Shovelomics: high throughput phenotyping of maize (Zea mays L.) root architecture in the field.Plant Soil34175–87. 10.1007/s11104-010-0623-8

  • CrossRef
  • Google Scholar

• 238

  TurraE. M.de OliveiraD. A. A.ValenteB. D.TeixeiraE.deA.PradoS.et al (2012). Estimation of genetic parameters for body weights of Nile tilapia Oreochromis niloticus using random regression models.Aquaculture354–35531–37. 10.1016/j.aquaculture.2012.04.035

  • CrossRef
  • Google Scholar

• 239

  van DusschotenD.MetznerR.KochsJ.PostmaJ. A.PflugfelderD.BuehlerJ.et al (2016). Quantitative 3D analysis of plant roots growing in soil using magnetic resonance imaging.Plant Physiol.17001388.2015. 10.1104/pp.15.01388

  • CrossRef
  • Google Scholar

• 240

  van EeuwijkF. A.Bustos-KortsD.MilletE. J.BoerM. P.KruijerW.ThompsonA.et al (2018). Modelling strategies for assessing and increasing the effectiveness of new phenotyping techniques in plant breeding.Plant Sci.28223–39. 10.1016/j.plantsci.2018.06.018

  • CrossRef
  • Google Scholar

• 241

  van PeltM. L.MeuwissenT. H. E.de JongG.VeerkampR. F. (2015). Genetic analysis of longevity in Dutch dairy cattle using random regression.J. Dairy Sci.984117–4130. 10.3168/jds.2014-9090

  • CrossRef
  • Google Scholar

• 242

  VanhataloJ.LiZ.SillanpääM. J. (2019). A Gaussian process model and bayesian variable selection for mapping function-valued quantitative traits with incomplete phenotypic data.Bioinformatics353684–3692. 10.1093/bioinformatics/btz164

  • CrossRef
  • Google Scholar

• 243

  VargasG.SchenkelF. S.BritoL. F.NevesH. H.deR.MunariD. P.et al (2018). Unravelling biological biotypes for growth, visual score and reproductive traits in Nellore cattle via principal component analysis.Livest. Sci.21737–43. 10.1016/j.livsci.2018.09.010

  • CrossRef
  • Google Scholar

• 244

  VaronaL.LegarraA.ToroM. A.VitezicaZ. G. (2018). Non-additive effects in genomic selection.Front. Genet.9:78. 10.3389/fgene.2018.00078

  • CrossRef
  • Google Scholar

• 245

  VelazcoJ. G.MalosettiM.HuntC. H.MaceE. S.JordanD. R.van EeuwijkF. A. (2019). Combining pedigree and genomic information to improve prediction quality: an example in sorghum.Theor. Appl. Genet.1322055–2067. 10.1007/s00122-019-03337-w

  • CrossRef
  • Google Scholar

• 246

  VernonA. J.AllisonJ. C. S. (1963). A method of calculating net assimilation rate.Nature200:814. 10.1038/200814a0

  • CrossRef
  • Google Scholar

• 247

  WadeK. M.QuaasR. L.Van VleckL. D. (1993). Estimation of the parameters involved in a first-order autoregressive process for contemporary groups.J. Dairy Sci.763033–3040. 10.3168/JDS.S0022-0302(93)77643-2

  • CrossRef
  • Google Scholar

• 248

  WangL. A.GoonewardeneZ. (2004). The use of MIXED models in the analysis of animal experiments with repeated measures data.Can. J. Anim. Sci.841–11. 10.4141/a03-123

  • CrossRef
  • Google Scholar

• 249

  WangX.ZhangR.SongW.HanL.LiuX.SunX.et al (2019). Dynamic plant height QTL revealed in maize through remote sensing phenotyping using a high-throughput unmanned aerial vehicle (UAV).Sci. Rep.9:3458. 10.1038/s41598-019-39448-z

  • CrossRef
  • Google Scholar

• 250

  WhalleyW. R.BinleyA.WattsC. W.ShanahanP.DoddI. C.OberE. S.et al (2017). Methods to estimate changes in soil water for phenotyping root activity in the field.Plant Soil415407–422. 10.1007/s11104-016-3161-1

  • CrossRef
  • Google Scholar

• 251

  WolfingerR. D. (1996). Heterogeneous variance-covariance structures for repeated measures.J. Agric. Biol. Environ. Stat.1205–230.

  • Google Scholar

• 252

  WuD.GuoZ.YeJ.FengH.LiuJ.ChenG.et al (2019). Combining high-throughput micro-CT-RGB phenotyping and genome-wide association study to dissect the genetic architecture of tiller growth in rice.J. Exp. Bot.70545–561. 10.1093/jxb/ery373

  • CrossRef
  • Google Scholar

• 253

  WuR.MaC. X.LinM.WangZ.CasellaG. (2004). Functional mapping of quantitative trait loci underlying growth trajectories using a transform-both-sides logistic model.Biometrics60729–738. 10.1111/j.0006-341X.2004.00223.x

  • CrossRef
  • Google Scholar

• 254

  WuR. L.LinM. (2006). Functional mapping – How to map and study the genetic architecture of dynamic complex traits.Nat. Rev. Genet.7229–237. 10.1038/nrg1804

  • CrossRef
  • Google Scholar

• 255

  WürschumT.LiuW.BusemeyerL.TuckerM. R.ReifJ. C.WeissmannE. A.et al (2014). Mapping dynamic QTL for plant height in triticale.BMC Genet.15:59. 10.1186/1471-2156-15-59

  • CrossRef
  • Google Scholar

• 256

  XavierA.HallB.HearstA. A.CherkauerK. A.RaineyK. M. (2017). Genetic architecture of phenomic-enabled canopy coverage in Glycine max.Genetics2061081–1089. 10.1534/genetics.116.198713

  • CrossRef
  • Google Scholar

• 257

  XuR.LiC.PatersonA. H. (2019). Multispectral imaging and unmanned aerial systems for cotton plant phenotyping.PLoS ONE14:e0205083. 10.1371/journal.pone.0205083

  • CrossRef
  • Google Scholar

• 258

  YanW.RajcanI. (2003). Prediction of cultivar performance based on single- versus multiple-year tests in soybean.Crop Sci.43549–555. 10.2135/CROPSCI2003.5490

  • CrossRef
  • Google Scholar

• 259

  YangG.LiuJ.ZhaoC.LiZ. Z.HuangY.YuH.et al (2017). Unmanned aerial vehicle remote sensing for field-based crop phenotyping: current status and perspectives.Front. Plant Sci.8:1111. 10.3389/fpls.2017.01111

  • CrossRef
  • Google Scholar

• 260

  YangR.TianQ.XuS. (2006). Mapping quantitative trait loci for longitudinal traits in line crosses.Genetics1732339–2356. 10.1534/genetics.105.054775

  • CrossRef
  • Google Scholar

• 261

  YangR.XuS. (2007). Bayesian shrinkage analysis of quantitative trait Loci for dynamic traits.Genetics1761169–1185. 10.1534/genetics.106.064279

  • CrossRef
  • Google Scholar

• 262

  YanoK.MorinakaY.WangF.HuangP.TakeharaS.HiraiT.et al (2019). GWAS with principal component analysis identifies a gene comprehensively controlling rice architecture.Proc. Natl. Acad. Sci. U.S.A.116:201904964. 10.1073/pnas.1904964116

  • CrossRef
  • Google Scholar

• 263

  YinX.GoudriaanJ.LantingaE. A.VosJ.SpiertzH. J. (2003). A flexible sigmoid function of determinate growth.Ann. Bot.91361–371. 10.1093/aob/mcg029

  • CrossRef
  • Google Scholar

• 264

  YorkL. M.SlackS.BennettM. J.FoulkesM. J. (2018). Wheat shovelomics I: a field phenotyping approach for characterising the structure and function of root systems in tillering species.bioRxiv[preprint]10.1101/280875

  • CrossRef
  • Google Scholar

• 265

  ZhangW.GaoX.ShiX.ZhuB.WangZ.GaoH.et al (2018). PCA-based multiple-trait GWAS analysis: a powerful model for exploring pleiotropy.Animals (Basel)8:239. 10.3390/ani8120239

  • CrossRef
  • Google Scholar

• 266

  ZhangX.HuangC.WuD.QiaoF.LiW.DuanL.et al (2017). High-throughput phenotyping and QTL mapping reveals the genetic architecture of maize plant growth.Plant Physiol.1731554–1564. 10.1104/pp.16.01516

  • CrossRef
  • Google Scholar

• 267

  ZhaoC.ZhangY.DuJ.GuoX.WenW.GuS.et al (2019). Crop phenomics: current status and perspectives.Front. Plant Sci.10:714. 10.3389/fpls.2019.00714

  • CrossRef
  • Google Scholar

• 268

  ZhongZ.SunA. Y.YangQ.OuyangQ. (2019). A deep learning approach to anomaly detection in geological carbon sequestration sites using pressure measurements.J. Hydrol.573885–894. 10.1016/j.jhydrol.2019.04.015

  • CrossRef
  • Google Scholar

• 269

  ZouJ.HussM.AbidA.MohammadiP.TorkamaniA.TelentiA. (2019). A primer on deep learning in genomics.Nat. Genet.5112–18. 10.1038/s41588-018-0295-5

  • CrossRef
  • Google Scholar