Volume 14, Issue 3 e20160

SPECIAL ISSUE: ADVANCES IN GENOMIC SELECTION AND APPLICATION OF MACHINE LEARNING IN GENOMIC PREDICTION

Open Access

Comprehensive analytical and empirical evaluation of genomic prediction across diverse accessions in maize

Matthew J. Dzievit

orcid.org/0000-0002-1437-1027

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing - original draft, Writing - review & editing

Search for more papers by this author

Tingting Guo,

Tingting Guo

orcid.org/0000-0002-6647-6998

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Methodology, Resources, Writing - review & editing

Search for more papers by this author

Xianran Li,

Xianran Li

orcid.org/0000-0002-4252-6911

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Methodology, Resources, Writing - review & editing

Search for more papers by this author

Jianming Yu,

Corresponding Author

Jianming Yu

[email protected]

orcid.org/0000-0001-5326-3099

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Correspondence

Jianming Yu, Dep. of Agronomy, Iowa State Univ., 2104 Agronomy Hall, Ames, IA 50011–1051.

Email: [email protected]

Contribution: Conceptualization, Funding acquisition, Methodology, Project administration, Writing - review & editing

Search for more papers by this author

Matthew J. Dzievit,

Matthew J. Dzievit

orcid.org/0000-0002-1437-1027

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing - original draft, Writing - review & editing

Search for more papers by this author

Tingting Guo,

Tingting Guo

orcid.org/0000-0002-6647-6998

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Methodology, Resources, Writing - review & editing

Search for more papers by this author

Xianran Li,

Xianran Li

orcid.org/0000-0002-4252-6911

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Contribution: Methodology, Resources, Writing - review & editing

Search for more papers by this author

Jianming Yu,

Corresponding Author

Jianming Yu

[email protected]

orcid.org/0000-0001-5326-3099

Dep. of Agronomy, Iowa State Univ., Ames, IA, 50011 USA

Correspondence

Jianming Yu, Dep. of Agronomy, Iowa State Univ., 2104 Agronomy Hall, Ames, IA 50011–1051.

Email: [email protected]

Contribution: Conceptualization, Funding acquisition, Methodology, Project administration, Writing - review & editing

Search for more papers by this author

First published: 18 October 2021

https://doi.org/10.1002/tpg2.20160

Citations: 3

Assigned to Associate Editor Rajeev Varshney.

Share a link

Email
Facebook
x
LinkedIn
Reddit
Wechat

Abstract

Efficiently exploiting natural genetic diversity captured by accessions stored in genebanks is crucial to genetic improvement of major crops. Selecting accessions of interest from genebanks has traditionally required information from extensive and expensive evaluation; however, low-cost genotyping combined with genomic prediction have enabled us to generate predicted genetic merits for the entire set with targeted phenotypic evaluation of representative subsets. To explore this general approach, analytical assessment and empirical validation of the maize (Zea mays L.) association population (MAP) as a training population were conducted in the present study. Cross-validation within the MAP revealed mostly modest to strong predictive ability for 36 traits, generally in parallel with the square root of heritability. The MAP was then used to train the prediction models to generate genomic estimated breeding values (GEBVs) for the Ames Diversity Panel. Empirical validation conducted for nine traits across two validation populations confirmed the accuracy level indicated by the cross-validation of the training population. An upper bound for reliability (U value) was calculated for the accessions in the prediction population using genotypic data. The group of accessions with high U values generally had high predictive ability, even though the range of observed trait values was similar to the group of accessions with low U values. Our comprehensive analysis validated the general approach of turbocharging genebanks with genomics and genomic prediction. In addition, breeders and researchers can consider both GEBVs and U values to balance the needs of improving specific traits and broadening genetic diversity when selecting accessions from genebanks.

Abbreviations

AmesDP: Ames Diversity Panel
CV: cross-validation
EH: EarHeight
GBS: genotyping by sequencing
GEBV: genomic estimated breeding value
MAP: Maize Association Population
MLA: MiddleLeafAngle
PH: PlantHeight
SNP: single nucleotide polymorphism
ULA: UpperLeafAngle.

1 INTRODUCTION

Modern plant breeding is like a double-edged sword: intensive selection coupled with modified management practices produces high-yielding cultivars but at the same time narrows the genetic bases of the elite breeding germplasm pools (Tanksley & McCouch, 1997). This reduction of genetic diversity in our major crops increases their vulnerability to the ever-changing climatic conditions and rapid evolution of diseases and insects. Private and public breeders have recognized the need to expand our elite germplasm's genetic diversity to safeguard against these future challenges for some time (Goodman, 2005; Mikel & Dudley, 2006), but we have only scratched the surface in integrating available diversity into our elite germplasm (Fox et al., 2015; Tanksley et al., 1996; Xiao et al., 1998). Genebank curators have excelled in conserving the rich natural genetic diversity by preserving millions of plant accessions in genebanks across the world (FAO, 2010). Programs such as the Latin American Maize Project (Salhuana et al., 1997) and Germplasm Enhancement of Maize (Pollak, 2001) pre-breed diverse genebank accessions to increase the genetic diversity of maize (Zea mays L.) germplasm available for breeders. Using any genetic materials from genebanks requires efficient determination of the accessions worth evaluating, which depends on a breeding or research program's objectives. Accomplishing this task begins with carefully curated passport, genotypic, and phenotypic data for the millions of genebank accessions (McCouch et al., 2012), yet obtaining this information has proven to be both expensive and time consuming (Houle et al., 2010).

Advances in genomic technologies, such as genotyping by sequencing (GBS) (Elshire et al., 2011), have helped overcome one of these obstacles by providing cheap and abundant single nucleotide polymorphisms (SNPs). Recently, the application of this technology in genebanks has provided large-scale genetic characterization of these collections (Crossa et al., 2016; Romay et al., 2013; Yu et al., 2016). Even with advancements in high-through phenotyping (e.g., Crain et al., 2018; Pugh et al., 2017; Salas Fernandez et al., 2017; Wang et al., 2021), large-scale phenotyping of genebank collections for important traits is still challenging. Instead, researchers have focused on extensive phenotypic characterization of manageable representative subsets of large genebank collections, commonly called “association panels” (e.g., Flint-Garcia et al., 2005), “mini-core collections” (e.g., Upadhyaya et al., 2009), or “diversity panels” (e.g., Zhao et al., 2011).

Genomic prediction (Bernardo & Yu, 2007; Crossa et al., 2017; Heffner et al., 2009; Meuwissen et al., 2001) exploits low-cost genotypic data to bridge the gap between phenotypic data collected from the training set to the unphenotyped population of potential selection candidates. When the genotypically and phenotypically characterized association panels are used to train a prediction model and generate genomic estimated breeding values (GEBVs) for the untested but genotyped genebank collections, this approach is termed “turbocharging genebanks” (Yu et al., 2016). Research to uncover the hidden potential in genebanks across the world has been reported in sorghum [Sorghum bicolor (L.) Moench] (Yu et al., 2016), soybean [Glycine max (L.) Merr.] (Jarquin et al., 2016), wheat (Triticum aestivum L.) (Crossa et al., 2016), and cauliflower (Brassica oleracea var. botrytis) (Thorwarth et al., 2018). Although genomic prediction involving the diverse maize accessions has been conducted through cross-validation (CV) for plant height and flowering time (Peiffer et al., 2013, 2014) and seedling root length (Pace et al., 2015), a large-scale genomic prediction analysis of the Ames Diversity Panel (AmesDP) with an extensively phenotyped association panel has not yet been conducted.

Core Ideas

Genetic diversity captured in accessions stored in genebanks is critical to crop improvement.
Genomics and genomic prediction can be leveraged for efficient mining of genebank germplasm.
Maize Association Population and Ames Diversity Panel were used to assess predictive ability.
Empirical validation confirmed the moderate to high predictive ability from cross-validation.
Selection of accessions can consider both predicted trait values and reliability values.

The AmesDP is a collection of 2,537 unique accessions stored at the USDA-ARS North Central Regional Plant Introduction Station in Ames, IA (Romay et al., 2013). Within this collection are maize landraces developed by farmers and inbreds from public breeding programs from around the world. Additionally, it contains recently expired plant variety protection inbreds registered by private companies. The Maize Association Population (MAP, or Goodman association panel) is a collection of diverse inbreds from public breeding programs from around the world (Flint-Garcia et al., 2005). The accessions from the AmesDP and MAP were genotyped using GBS, and the MAP was found to represent 75% of the total allelic diversity of the AmesDP (Romay et al., 2013). Both panels also have extensive phenotypic data available because the MAP was previously phenotyped for numerous traits across a wide range of environments, and the AmesDP was phenotyped for seven traits related to height and flowering time (Hung et al., 2012; Peiffer et al., 2013, 2014; Romay et al., 2013). Together, these datasets provide an excellent opportunity to explore the prediction potential of the MAP across a range of traits and to predict and validate GEBVs for the AmesDP.

In this study, we first filtered and imputed previously obtained SNPs for the MAP and the AmesDP. The MAP was used to train the prediction model and generate GEBVs for the AmesDP across 36 traits (Figure 1a,b). In addition to conducting CV, we conducted empirical validation of predictive ability for four traits using 292 randomly selected accessions from the AmesDP (Figure 1c) and for seven traits previously phenotyped (Figure 1b) for a 2,043-accession set. Finally, we assessed the effect of genetic relationship on predictive ability by examining the upper bound for reliability (U value) for nine traits to gain insights into how to better leverage genomics and genomic prediction for trait improvement and genetic diversity exploration.

Details are in the caption following the image — **FIGURE 1**
Open in figure viewer

Genomic prediction and validation outline. (a) Maize association population was used to train a genomic prediction model and produce genomic estimated breeding values for the Ames diversity panel. A small empirical validation population and a large empirical validation population were examined for different traits. Reference population includes the training population and the prediction population. (b) The set of 36 traits belonging to six different trait types were studied. *Seven traits studied for the large empirical validation population. (c) Four traits studied for the small empirical validation population. GDD, growing degree days; NIR, near-infrared reflectance

2 MATERIALS AND METHODS

2.1 Genetic materials

Accessions from the MAP (Flint-Garcia et al., 2005) and AmesDP (Romay et al., 2013) were used in the present study. The 281 MAP accessions constituted the training population, and the 2,537 AmesDP accessions, not overlapping with the MAP, constituted the prediction population (Figure 1a; Supplemental Table S1). Together, these two sets of accessions formed the 2,818-accession reference population (Figure 2). From the prediction population, two empirical validation populations were used: (a) 292 randomly selected accessions as the small empirical validation population and (b) 2,043 previously phenotyped accessions (Peiffer et al., 2013, 2014; Romay et al., 2013) as the large empirical validation population (Supplemental Table S1).

2.2 Genotypic data and imputation

The ZeaGBSv27 (Glaubitz et al., 2014; panzea.org) SNP set was previously generated by GBS and contains 954,882 SNPs (AGPv3). A detailed workflow of the scripts and options chosen was used to process the genotypic data. Briefly, the raw SNP file was converted to variant call format using the command line options in TASSEL (Bradbury et al., 2007). An in-house python script was used to combine samples that were genotyped multiple times (Supplemental Table S1) and to split the reference population into the 281-accesion training population and 2,537-accession prediction population.

Different filtering conditions were applied to both populations, and the overlapping SNPs were extracted from each population. The program ‘VCFtools v0.1.15’ (VCFtools) (Danecek et al., 2011) was used to split the two variant call format files containing the training and prediction genotypes into 10 maize chromosomes. Filtering in the training population excluded indels and SNPs that were monomorphic, had >80% missing data, and had a minor allele frequency ≤1%. For the prediction population, similar filtering criteria were used, except we did not filter for minor allele frequency. After filtering, the remaining 344,477 SNPs were imputed for missing data using ‘Beagle, v4.1’ with program defaults (Browning & Browning, 2007). The SNPs that overlapped between the two populations were extracted via VCFtools. The training and prediction populations were indexed using ‘tabix, v1.6’ (Li, 2011) and then merged with VCFtools.

2.3 Genetic diversity

Population structure of the reference population was estimated using principal component analysis and ‘ADMIXTURE, v1.3’ (ADMIXTURE) (Alexander et al., 2009). The 344,477 filtered SNPs were used in a principal component analysis via ‘plink v1.9’ (Chang et al., 2015; Purcell et al., 2007) to assist in identifying the number of subpopulations. To prepare the filtered SNPs for ADMIXTURE, ‘plink v1.9’ was used to prune SNPs in linkage disequilibrium with the following options: 50-SNP window, a step size of 10, and a r² threshold of .20. The remaining 103,408 SNPs were used in a CV ADMIXTURE procedure (k = 1…10). Results from the CV, principal component analysis, and previous subpopulation classifications (Romay et al., 2013) were considered together to classify five subpopulations (Figure 2). Individual with membership coefficients ≥70% were assigned to the respective subpopulations, and the rest were assigned to a mixed group.

2.4 Phenotypes for training and empirical validation populations

We used a mixed model to analyze previous phenotypic data (Hung et al., 2012) for the 281-accession training population accessed from panzea.org. A pipeline implemented in R was used to analyze the training population phenotypic data. This dataset originally contained 53 traits that were phenotyped across 16 environments (year and location combinations); not all traits were evaluated across every environment (Hung et al., 2012). They were classified into six trait types: plant architecture, flowering, tassel, ear, kernel, and other (Figure 1a). Harmonic means (Holland et al., 2010) were calculated for each trait to determine the average number of environments per individual. Traits involving “border plants” or with a harmonic mean <1.5 were removed. An additional trait, “TotalKernelWeight,” was calculated as the difference between “EarWeight” and “CobWeight.” For the 36 traits chosen for further analysis, genotypes were treated as random effects, and environments were treated as fixed effects (Möhring & Piepho, 2009) to obtain genotypic values for each individual using the R package ‘lme4 v1.1-17’ (Bates et al., 2015). Variance components were estimated using lme4 to calculate broad-sense heritability on an entry-mean basis with the following equation:

H^{2} = \frac{{\hat{σ}}_{g}^{2}}{{\hat{σ}}_{g}^{2} + \frac{{\hat{σ}}_{e}^{2}}{e_{h}}}

where

{\hat{σ}}_{g}^{2}

is the estimated genetic variance,

{\hat{σ}}_{e}^{2}

is the estimated error variance, and e_h is the harmonic mean for the environments (Holland et al., 2010; Piepho & Möhring, 2007).

The 292-accession small empirical validation population was phenotyped for “PlantHeight” (PH), “EarHeight” (EH), “UpperLeafAngle” (ULA), and ”MiddleLeafAngle” (MLA) (Figure 1c). Leaf angle measurements were taken from digital images (Dzievit et al., 2018) by measuring the angle between horizontal and the middle of the midrib. The trait ULA (°) is defined as the leaf immediately below the upper most leaf, or flag leaf, and MLA (°) is defined as the second leaf below the ear leaf. The trait PH (cm) was measured as the distance between the soil surface and base of the flag leaf, and EH (cm) was measured as the distance between the soil surface and the ear-bearing node of the uppermost ear. The small empirical population was planted in Boone, IA, in 2015, 2016, and 2017, with a single replication. Genotypic values for the 292-accession small empirical validation population and the combined training and empirical validation population were determined using the previously described method.

Trait data for the large 2,043-accession empirical validation population were obtained from panzea.org (accessed Sept. 2018) for seven traits: PH; EH; “DaysToSilk”; “GDDDaysToSilk,” where GDD is growing degree days; “DaysToTassel”; “GDDDaysToTassel”; and “GDDAnthesis-SilkingInterval” (Peiffer et al., 2013, 2014; Romay et al., 2013). In total, nine traits were validated across the two empirical validation populations, with PH and EH overlapping between the two.

2.5 Genomic prediction and upper bound for reliability

The GEBVs for 36 traits were generated for each accession in the prediction population. The R package ‘rr-BLUP v4.6’ (Endelman, 2011) was used to implement a ridge-regression best linear unbiased prediction (RR-BLUP) model (Endelman, 2011; Whittaker et al., 2000) using the 344,477 SNPs and phenotypic data for the MAP. The resulting model and genotypic data of the prediction population were combined to generate GEBVs for 36 traits. A reverse prediction was assessed following the same procedure, where data from the 292-accession small empirical validation population were used to estimate marker effects and generate GEBVs for ULA, MLA, PH, and EH for the 281-accession training population.

A k-fold CV method was applied to the MAP to evaluate predictive ability for all 36 traits. The Pearson correlation (r) of GEBVs and observed values were used to estimate predictive ability. Three fold sizes (k) were tested (k = 2, 5, and 10), and predictive ability was averaged across 50 iterations for each trait and k-fold combination. A single iteration split the dataset into approximately equal size k-folds, where one fold was to be predicted and the remaining folds were the training population for the prediction model. This process was repeated until every fold was predicted, and the correlation between observed values and predictions for that iteration was determined. The means of all 50 iterations were reported.

The U value (Karaman et al., 2016) was calculated for all accessions in the prediction population using an in-house R script and the training and prediction population's genotypic data. The U value was obtained as $U = {\hat{v}}^{'} \hat{v} / \hat{v} v$ , where $\hat{v} = M^{'} {({MM}^{'})}^{- 1} Mv$ , M is a matrix of training set genotypes, and v is a vector of validation genotypes (Karaman et al., 2016). This assesses the representation of the prediction population within all linear combinations of the genotypes in the training population. The U values were ranked from highest to lowest. To compare predictive ability within each group, the small empirical validation population was separated into three groups (Top50, Middle, and Bottom50), and the large empirical validation population was separated into three groups (Top400, Middle, and Bottom400).

3 RESULTS

3.1 Genetic relationships from markers

The genotypic data for the 281-accesion training population and the 2,537-accession prediction population were imputed and filtered. The training and prediction populations were filtered first and then combined to obtain the set of 344,477 common SNPs that were used for imputation. Population structure was determined using multiple methods, including ancestry estimation, principal component analysis, and previous research. From the 344,477 common SNPs in the reference population, 30% (103,408) were retained after removing other SNPs in linkage disequilibrium. Examining findings from CV error analysis (Supplemental Figure S1), principal component analysis (Figure 2), and results from an earlier study (Romay et al., 2013), we determined five subpopulations: stiff stalk, non-stiff stalk, tropical, popcorn, and sweetcorn. A membership cutoff value of 70% resulted in 57% of the accessions in the reference population assigned to a single subpopulation; the rest were assigned to a mixed group (Supplemental Table S1).

The SNP data for the training and prediction populations were used to calculate a U value (Karaman et al., 2016) for each individual within the prediction population (Supplemental Table S1). The U value is a numerical estimator of the individual's genetic representation within the training population. The obtained U values ranged from 0.507 to 0.991, with an average value of 0.653 and a median value of 0.608 (Supplemental Table S2). Within subpopulations, average U value estimates ranged from 0.587 in the tropical population to 0.824 in the stiff stalk population. Distributions across and within each subpopulation were mostly right-skewed, reflecting the overall compositions of the prediction population and the training population (Supplemental Figure S2).

3.2 Heritability estimates

Broad-sense heritability and genotypic values were calculated for 36 traits in the MAP (Supplemental Table S3). Harmonic means for each trait were estimated to indicate the number of environments in which traits were phenotyped and ranged from 1.81 for “Spikelets-PrimaryBranch” to 9.07 for PH (Supplemental Table S4). Using these estimates and variance component estimates, we calculated broad-sense heritability on an entry-mean basis for the 36 traits that ranged from 0.18 for “TilleringIndex” to 0.97 for “DaysToTassel” (Supplemental Table S4).

For the 292-accession small validation population, phenotypic distributions for all four traits (ULA, MLA, PH, and EH) were normally distributed (Supplemental Figure S3). Pearson correlations between years ranged from 0.70 to 0.78 (Supplemental Figure S4). Heritability on an entry-mean basis ranged from 0.88 for ULA to 0.91 for PH (Supplemental Table S4). For the large empirical validation population, heritability ranged from 0.86 to 0.92. To the best of our knowledge, previous studies only reported heritability values for PH, EH, “DaysToTassel,” and “DaystoAnthesis” (Peiffer et al., 2013, 2014).

When we combined the small empirical validation population with the training population MAP, harmonic means in the 573-accession became 2.65 for MLA to 4.42 for PH (Supplemental Table S4). Heritability on an entry-mean basis was calculated again for these four traits and ranged from 0.83 for MLA to 0.91 for EH and PH (Supplemental Table S4).

3.3 Cross-validation of training data

Cross-validation with three fold sizes (2, 5, and 10) was used to assess the MAP's predictive ability across 36 traits. As the size of the training population increased, predictive ability increased, whereas its standard deviation decreased for all traits (Supplemental Figure 5; Supplemental Table S4). In the 10-fold CV, average predictive ability across the 50 iterations ranged from 0.08 for “StandCount” to 0.88 for “GDDDaystoTassel” (Figure 3). Overall, we observed that predictive ability is significantly correlated (r = .79; p < .01) (Supplemental Figures S6 and S7) with the square root of heritability.

To further assess the impact of sample size on predictive ability and data quality obtained for a small validation population, we conducted additional CV analysis with data from the small empirical population alone and the combined population of MAP and four traits (ULA, MLA, PH, and EH) within the small empirical population. The combined population had the highest average predictive ability (Supplemental Figure S8) and the lowest SD (Supplemental Table S4). With the 10-fold CV, average predictive ability for the small empirical population was higher for PH, lower for ULA and MLA, and similar for EH when compared with that of the MAP (Supplemental Figure S8).

3.4 Validating predictions and relationship with upper bound for reliability

With the MAP training population, predictions were made for the 2,537-accession prediction population (Supplemental Table S5). The experiment using a 292-accession small empirical validation panel was conducted to empirically validate predictions for ULA, MLA, PH, and EH. The predictive ability for the four traits measured in the empirical validation experiment was generally equal to or lower than the average predictive ability observed from the 10-fold CV of the MAP, the small validation population, and combined populations (Supplemental Figure S8). The predictive ability was moderate and mostly consistent for all four traits: 0.56 for ULA, 0.40 for MLA, 0.57 for PH, and 0.65 for EH (Figure 4). On the other hand, within-subpopulation predictive ability varied for all four traits. For ULA, it ranged from 0.13 for popcorn to 0.78 for stiff stalk; for MLA, from −0.66 for sweet corn to 0.45 for popcorn; for PH, from 0.30 for stiff stalk to 0.81 for popcorn; and for EH, from 0.19 for sweet corn to 0.72 for popcorn.

As an extra investigation, the 292-accession small empirical validation population was treated as the training population to generate GEBVs for the 281-accession MAP for ULA, MLA, PH, and EH to assess its predictive ability. Predictive ability for this reverse prediction was generally similar to the forward prediction (Supplemental Figure S9), except we observed a higher predictive ability for ULA (0.67). The number of individuals within a subpopulation was different in this training population versus the MAP, and within-subpopulation predictive ability varied as well.

For the 2,043-accession large prediction population, we compared the GEBVs generated from the MAP training population with the observed trait values reported in two previous studies (Peiffer et al., 2013, 2014) for seven traits: PH, EH, “DaystoSilk”; “GDDDaystoSilk,” “DaysToTassel,” “GDDDaystoTassel,” and “GDDAnthesis-SilkingInterval.” Predictive ability for these seven traits ranged from 0.04 for “GDDAnthesis-SilkingInterval,” which is a derived trait, to 0.86 for “GDDDaystoTassel” (Supplemental Figure S10). Within-subpopulation predictions also varied widely for these seven traits (Supplemental Figure S10). Across all traits except “GDDAnthesis-SilkingInterval,” the mixed subpopulation had the highest predictive ability. Excluding the mixed subpopulation, predictive ability within the non-stiff stalk population tended to be the highest across most of the seven traits, whereas the popcorn subpopulation tended to have the lowest predictive ability.

Using only the genotypic data, U value was calculated for the 2,537-accession prediction population. Most of the predicted values centering on the trait mean had lower U values, whereas those with extreme (lowest and highest) GEBVs had higher U values (Supplemental Figure S11). This formed a distinct “U” shape that was consistent across all 36 traits, indicating that for accessions with low U values and less relation to the training population, the model generated GEBVs close to the trait mean. In addition, accessions classified into different subpopulations had their unique positions on the plot of U value versus GEBV, which was a result of the genetic composition of the training population, observed trait values of the training population, and genetic composition of the prediction population. For the 292-accession empirical validation population, accessions were ranked by U value from highest to lowest and classified into three groups (Top50, Middle, and Bottom50) to compare predictive ability among the different groups. Accessions from the mixed and tropical subpopulation composed most of the accessions in the Bottom50 (Figure 5). Across all four traits (ULA, MLA, PH, and EH), the top 50 group had a higher predictive ability than the overall, and the Bottom50 group was lower. Predictive ability for accessions in the Middle group was similar to the overall predictive ability. For the 2,043-accession large empirical validation set, the difference in predictive ability between the Top400 group and Bottom400 group was observed for four of the seven traits (Supplemental Figure 12).

4 DISCUSSION

Expanding the diversity of our elite breeding germplasm will rely heavily on effective mining of useful alleles from accessions warehoused in genebanks across the world. Most genebanks have focused primarily on conserving accessions rather than genotypic and phenotypic characterization and effective usage of this material. Fortunately, high-throughput sequencing and genotyping combined with increased investments in extensive phenotyping of representative subsets provide an opportunity for leveraging genomic prediction to exploit the genetic diversity housed in genebanks (Mascher et al., 2019; McCouch et al., 2012; Yu et al., 2016, 2020). It is desirable to keep generating additional information to both demonstrate the utility of this approach and identify areas to improve.

In maize, extensive phenotyping of the MAP for numerous types of traits across multiple environments (Hung et al., 2012) and investments in dense genotypic data across the AmesDP (Romay et al., 2013) provided such an opportunity. The 36 traits investigated in this study include different components of maize development, such as plant architecture, flowering, tassel, ear, and kernel. Across all traits, predictive ability increased with training population size, consistent with previous findings (Crossa et al., 2016; Muleta et al., 2017; Yu et al., 2016). Furthermore, predictive ability was strongly correlated with the square root of heritability, highlighting the need for increasing phenotyping accuracies through improved experimental designs and balancing traditional versus precision-phenotyping methods (Cobb et al., 2013). For example, high-throughput phenotyping methods using unoccupied aerial vehicles (e.g., Crain et al., 2017; Pugh et al., 2017; Wang et al., 2021) could be deployed for quick and highly accurate phenotyping for traits like plant height with a single replication across locations, whereas traits like kernel volume may benefit from improved ear and kernel sampling and phenotyping methods to increase heritability and therefore predictive ability.

Cross-validation is a resampling procedure used to evaluate a training population's predictive ability for a given trait. Validating predictions with empirical experiments would provide additional support to help convince breeders or researchers to leverage predicted trait values in selecting material from genebanks. Accordingly, we conducted experiments to measure four traits (ULA, MLA, PH, and EH) across 292 accessions from the AmesDP, and the results are encouraging. Predictive ability obtained for empirical validation was generally lower than that from 10-fold cross-validation of the training population. Although phenotyping of the training population occurred across multiple geographical locations across the United States (Hung et al., 2012), the small empirical validation population was only evaluated in a single geographic location across 3 years. In addition, future research can consider incorporating environmental dimension to generate a series of specific prediction values for the environments used in validating experiments (Li et al., 2021).

Additionally, we investigated predictive ability with previously generated phenotypic data for seven traits (Peiffer et al., 2013, 2014; Romay et al., 2013), two of which overlap with our collected data (PH and EH). These seven traits were evaluated across multiple environments and include a larger number of the accessions. Predictive ability for these seven traits was close to the average predictive ability observed in the 10-fold cross validation of the training population. Together, these results support the predictions made with the MAP as a training population for the AmesDP.

It is well established that genetic relationship between training and prediction population affects predictive ability (Crossa et al., 2017). In this study, we investigated how U values affect predictive ability. Across nine traits, predictive ability for the group of accessions with high U values was generally higher than for the group of accessions with low U values. Our findings are consistent with the original concept and simulations (Karaman et al., 2016) and empirical studies (Yu et al., 2016, 2020). At the same time, the genetic composition of the training population and the prediction population affected both the GEBVs and the U values. These results suggest that designing training populations to optimally increase U values of the prediction population would improve predictive ability and that visualizing the genetic composition of the prediction population together with GEBVs and U values is helpful to understand the process and the outcome. For example, tropical accessions and sweetcorn accessions had more extreme observed and predicted trait values as a group for several traits directly related to maturity or influenced by maturity.

Phenotypically evaluating numerous traits for all accessions in a genebank is prohibitively expensive and time consuming, particularly considering geographical adaptation zones and changing biotic and abiotic conditions. In addition to GEBVs, U values can help select accession from genebanks. The U value is derived directly from genome-wide marker data to quantify the estimability of genomic make-up of a selection candidate from genomic data of the training set (Karaman et al., 2016). It was clear that certain maize subpopulations had higher representation in the training population, leading to generally higher U values for selection accessions from those subpopulations. By checking both predictions and U values, breeders and researchers can have more confidence in the predicted trait values for a given accession with high U values, particularly if the focus is to simply select accessions with high trait values. On the other hand, if expanding diversity beyond what is captured in the training population is needed, some accessions with low U values but modest prediction values may also be selected. This is to capture diversity that is not well represented in the training population as a step-wise expansion strategy (Yu et al., 2020).

For many crop species in a genebank, it may be desirable to establish multiple training populations with different genetic compositions so that they can be evaluated at different sets of environments and adaptation zones. Besides germplasm information, different methods in training set design (Akdemir et al., 2015; Guo et al., 2019; He et al., 2016; Rincent et al., 2012, 2017) can be applied to the genomic information obtained for the collections (Mascher et al., 2019; McCouch et al., 2012; Yu et al., 2016) to optimize the process of establishing the initial training populations. Then, plots of reliability versus prediction for different traits should be investigated together with the diversity analysis to determine the validation set design. The whole process should follow the well-established understanding of performance evaluation, genetic diversity, and environmental heterogeneity as well as the concept of target population of genotypes and target population environments. Similar to many other genomic-assisted breeding research (Varshney et al., 2021), the emphasis of design will be critical, and coordinated research efforts toward turbocharging genebanks are needed to further improve the overall system to realize the potential of the genetic treasures.

ACKNOWLEDGMENTS

This work was supported by the Agriculture and Food Research Initiative competitive grant (2017-67007-25942) from the USDA National Institute of Food and Agriculture, the Iowa State University Raymond F. Baker Center for Plant Breeding, the DuPont Pioneer Graduate Assistantship, and the Iowa State University Plant Sciences Institute.

AUTHOR CONTRIBUTIONS

Matthew J. Dzievit: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Visualization; Writing-original draft; and Writing-review & editing. Tingting Guo: Methodology; Resources; and Writing-review & editing. Xianran Li: Methodology; Resources; and Writing-review & editing. JY: Conceptualization; Funding acquisition; Methodology; Project administration; and Writing-review & editing.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Open Research

DATA AVAILABILITY STATEMENT

Data and code used in this study is uploaded in an online public repository: https://github.com/mdzievit/Genomic_Prediction

Supporting Information

REFERENCES

Akdemir, D., Sanchez, J. I., & Jannink, J.-L. (2015). Optimization of genomic selection training populations with a genetic algorithm. Genetics Selection Evolution, 47, 38.
10.1186/s12711-015-0116-6
PubMedWeb of Science®Google Scholar
Alexander, D. H., Novembre, J., & Lange, K. (2009). Fast model-based estimation of ancestry in unrelated individuals. Genome Research, 19, 1655–1664. https://doi.org/10.1101/gr.094052.109
10.1101/gr.094052.109
CASPubMedWeb of Science®Google Scholar
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67. https://doi.org/10.18637/jss.v067.i01
10.18637/jss.v067.i01
Web of Science®Google Scholar
Bernardo, R., & Yu, J. (2007). Prospects for genomewide selection for quantitative traits in maize. Crop Science, 47, 1082–1090. https://doi.org/10.2135/cropsci2006.11.0690
10.2135/cropsci2006.11.0690
Web of Science®Google Scholar
Bradbury, P. J., Zhang, Z., Kroon, D. E., Casstevens, T. M., Ramdoss, Y., & Buckler, E. S. (2007). TASSEL: Software for association mapping of complex traits in diverse samples. Bioinformatics, 23, 2633–2635. https://doi.org/10.1093/bioinformatics/btm308
10.1093/bioinformatics/btm308
CASPubMedWeb of Science®Google Scholar
Browning, S. R., & Browning, B. L. (2007). Rapid and accurate haplotype phasing and missing-data inference for whole-genome association studies by use of localized haplotype clustering. American Journal of Human Genetics, 81, 1084–1097. https://doi.org/10.1086/521987
10.1086/521987
CASPubMedWeb of Science®Google Scholar
Chang, C. C., Chow, C. C., Tellier, L. C. A. M., Vattikuti, S., Purcell, S. M., & Lee, J. J.(2015). Second-generation PLINK: Rising to the challenge of larger and richer datasets. Gigascience, 4, 7. https://doi.org/10.1186/s13742-015-0047-8
10.1186/s13742-015-0047-8
PubMedWeb of Science®Google Scholar
Cobb, J. N., DeClerck, G., Greenberg, A., Clark, R., & McCouch, S. (2013). Next-generation phenotyping: Requirements and strategies for enhancing our understanding of genotype-phenotype relationships and its relevance to crop improvement. Theoretical and Applied Genetics, 126, 867–887. https://doi.org/10.1007/s00122-013-2066-0
10.1007/s00122-013-2066-0
PubMedWeb of Science®Google Scholar
Crain, J., Mondal, S., Rutkoski, J., Singh, R. P., & Poland, J. (2018). Combining high-throughput phenotyping and genomic information to increase prediction and selection accuracy in wheat breeding. The Plant Genome, 11, 170043. https://doi.org/10.3835/plantgenome2017.05.0043
10.3835/plantgenome2017.05.0043
Web of Science®Google Scholar
Crain, J., Reynolds, M., & Poland, J. (2017). Utilizing high-throughput phenotypic data for improved phenotypic selection of stress-adaptive traits in wheat. Crop Science, 57, 648. https://doi.org/10.2135/cropsci2016.02.0135
10.2135/cropsci2016.02.0135
Google Scholar
Crossa, J., Jarquín, D., Franco, J., Pérez-Rodríguez, P., Burgueño, J., Saint-Pierre, C., Vikram, P., Sansaloni, C., Petroli, C., Akdemir, D., Sneller, C., Reynolds, M., Tattaris, M., Payne, T., Guzman, C., Peña, R. J., Wenzl, P., & Singh, S. (2016). Genomic prediction of gene bank wheat landraces. G3: Genes|Genomes|Genetics, 6, 1819–1834. https://doi.org/10.1534/g3.116.029637
10.1534/g3.116.029637
CASWeb of Science®Google Scholar
Crossa, J., Pérez-Rodríguez, P., Cuevas, J., Montesinos-López, O., Jarquín, D., de los Campos, G., Burgueño, J., Camacho-González, J. M., Pérez-Elizalde, S., Beyene, Y., Dreisigacker, S., Singh, R., Zhang, X., Gowda, M., Roorkiwal, M., Rutkoski, J., & Varshney, R. K. (2017). Genomic selection in plant breeding: Methods, models, and perspectives. Trends in Plant Science, 22, 961–975. https://doi.org/10.1016/j.tplants.2017.08.011
10.1016/j.tplants.2017.08.011
CASPubMedWeb of Science®Google Scholar
Danecek, P., Auton, A., Abecasis, G., Albers, C. A., Banks, E., DePristo, M. A., Handsaker, R. E., Lunter, G., Marth, G. T., Sherry, S. T., McVean, G., & Durbin, R. (2011). The variant call format and VCFtools. Bioinformatics, 27, 2156–2158. https://doi.org/10.1093/bioinformatics/btr330
10.1093/bioinformatics/btr330
CASPubMedWeb of Science®Google Scholar
Dzievit, M. J., Li, X., & Yu, J. (2018). Dissection of leaf angle variation in maize through genetic mapping and meta-analysis. The Plant Genome, 12, 180024. https://doi.org/10.3835/plantgenome2018.05.0024
10.3835/plantgenome2018.05.0024
Web of Science®Google Scholar
Elshire, R. J., Glaubitz, J. C., Sun, Q., Poland, J. A., Kawamoto, K., Buckler, E. S., & Mitchell, S. E. (2011). A robust, simple genotyping-by-sequencing (GBS) approach for high diversity species. PLOS ONE, 6, e19379. https://doi.org/10.1371/journal.pone.0019379
10.1371/journal.pone.0019379
CASPubMedWeb of Science®Google Scholar
Endelman, J. B. (2011). Ridge regression and other kernels for genomic selection with R package rrBLUP. The Plant Genome, 4(3). https://doi.org/10.3835/plantgenome2011.08.0024
10.3835/plantgenome2011.08.0024
Web of Science®Google Scholar
FAO. (2010). Second report on the state of the world's plant genetic resources for food and agriculture. FAO.
Google Scholar
Flint-Garcia, S. A., Thuillet, A. C., Yu, J., Pressoir, G., Romero, S. M., Mitchell, S. E., Doebley, J., Kresovich, S., Goodman, M. M., & Buckler, E. S. (2005). Maize association population: A high-resolution platform for quantitative trait locus dissection. The Plant Journal, 44, 1054–1064. https://doi.org/10.1111/j.1365-313X.2005.02591.x
10.1111/j.1365-313X.2005.02591.x
CASPubMedWeb of Science®Google Scholar
Fox, C. M., Cary, T. R., Nelson, R. L., & Diers, B. W.(2015). Confirmation of a seed yield QTL in soybean. Crop Science, 55, 992–998. https://doi.org/10.2135/cropsci2014.10.0688
10.2135/cropsci2014.10.0688
Web of Science®Google Scholar
Glaubitz, J. C., Casstevens, T. M., Lu, F., Harriman, J., Elshire, R. J., Sun, Q., & Buckler, E. S. (2014). TASSEL-GBS: A high capacity genotyping by sequencing analysis pipeline. PLOS ONE, 9, e90346. https://doi.org/10.1371/journal.pone.0090346
10.1371/journal.pone.0090346
PubMedWeb of Science®Google Scholar
Goodman, M. M. (2005). Broadening the U.S. maize germplasm base. Maydica, 50, 203–214
Web of Science®Google Scholar
Guo, T., Yu, X., Li, X., Zhang, H., Zhu, C., Flint-Garcia, S., McMullen, M. D., Holland, J. B., Szalma, S. J., Wisser, R. J., & Yu, J. (2019). Optimal designs for genomic selection in hybrid crops. Molecular Plant, 12, 390–401.
10.1016/j.molp.2018.12.022
CASPubMedWeb of Science®Google Scholar
He, S., Schulthess, A. W., Mirdita, V., Zhao, Y., Korzun, V., Bothe, R., Ebmeyer, E., Reif, J. C., & Jiang, Y. (2016). Genomic selection in a commercial winter wheat population. Theoretical and Applied Genetics, 129, 641–651.
10.1007/s00122-015-2655-1
CASPubMedWeb of Science®Google Scholar
Heffner, E. L., Sorrells, M. E., & Jannink, J. (2009). Genomic selection for crop improvement. Crop Science, 49, 1–12. https://doi.org/10.2135/cropsci2008.08.0512
10.2135/cropsci2008.08.0512
CASWeb of Science®Google Scholar
Holland, J. B., Nyquist, W. E., & Cervantes-Martínez, C. T. (2010). Estimating and interpreting heritability for plant breeding: An update. John Wiley & Sons.
Google Scholar
Houle, D., Govindaraju, D. R., & Omholt, S. (2010). Phenomics: The next challenge. Nature Reviews Genetics, 11, 855–866. https://doi.org/10.1038/nrg2897
10.1038/nrg2897
CASPubMedWeb of Science®Google Scholar
Hung, H.-Y., Browne, C., Guill, K., Coles, N., Eller, M., Garcia, A., Lepak, N., Melia-Hancock, S., Oropeza-Rosas, M., Salvo, S., Upadyayula, N., Buckler, E. S., Flint-Garcia, S., McMullen, M. D., Rocheford, T. R., & Holland, J. B. (2012). The relationship between parental genetic or phenotypic divergence and progeny variation in the maize nested association mapping population. Heredity, 108, 490–499. https://doi.org/10.1038/hdy.2011.103
10.1038/hdy.2011.103
CASPubMedWeb of Science®Google Scholar
Jarquin, D., Specht, J., & Lorenz, A. (2016). Prospects of genomic prediction in the USDA soybean germplasm collection: Historical data creates robust models for enhancing selection of accessions. G3: Genes|Genomes|Genetics, 6, 2329–2341. https://doi.org/10.1534/g3.116.031443
10.1534/g3.116.031443
Web of Science®Google Scholar
Karaman, E., Cheng, H., Firat, M. Z., Garrick, D. J., & Fernando, R. L. (2016). An upper bound for accuracy of prediction using GBLUP. PLOS ONE, 11, e0161054. https://doi.org/10.1371/journal.pone.0161054
10.1371/journal.pone.0161054
PubMedWeb of Science®Google Scholar
Li, H. (2011). Tabix: Fast retrieval of sequence features from generic TAB-delimited files. Bioinformatics, 27, 718–719. https://doi.org/10.1093/bioinformatics/btq671
10.1093/bioinformatics/btq671
CASPubMedWeb of Science®Google Scholar
Li, X., Guo, T., Wang, J., Bekele, W. A., Sukumaran, S., Vanous, A. E., McNellie, J. P., Cortes, L. T., Lopes, M. S., Lamkey, K., Westgate, M. E., McKay, J., Archontoulis, S. V., Reynolds, M. P., Tinker, N. A., Schnable, P. S., & Yu, J. (2021). An integrated framework reinstating the environmental dimension for GWAS and genomic selection in crops. Molecular Plant, 14, 874–887.
Google Scholar
Mascher, M., Schreiber, M., Scholz, U., Graner, A., Reif, J. C., & Stein, N. (2019). Genebank genomics bridges the gap between the conservation of crop diversity and plant breeding. Nature Genetics, 51, 1076–1081.
10.1038/s41588-019-0443-6
CASPubMedWeb of Science®Google Scholar
McCouch, S. R., McNally, K. L., Wang, W., & Sackville Hamilton, R. (2012). Genomics of gene banks: A case study in rice. American Journal of Botany, 99, 407–423. https://doi.org/10.3732/ajb.1100385
10.3732/ajb.1100385
PubMedWeb of Science®Google Scholar
Meuwissen, T. H. E., Hayes, B. J., & Goddard, M. E. (2001). Prediction of total genetic value using genome-wide dense marker maps. Genetics, 157, 1819–1829. https://doi.org/10.1093/genetics/157.4.1819
10.1093/genetics/157.4.1819
CASPubMedWeb of Science®Google Scholar
Mikel, M. A., & Dudley, J. W. (2006). Evolution of North American dent corn from public to proprietary germplasm. Crop Science, 46, 1193–1205. https://doi.org/10.2135/cropsci2005.10-0371
10.2135/cropsci2005.10-0371
Web of Science®Google Scholar
Möhring, J., & Piepho, H. P. (2009). Comparison of weighting in two-stage analysis of plant breeding trials. Crop Science, 49, 1977–1988. https://doi.org/10.2135/cropsci2009.02.0083
10.2135/cropsci2009.02.0083
Web of Science®Google Scholar
Muleta, K. T., Bulli, P., Zhang, Z., Chen, X., & Pumphrey, M. (2017). Unlocking diversity in germplasm collections via genomic selection: A case study based on quantitative adult plant resistance to stripe rust in spring wheat. The Plant Genome, 10(3). https://doi.org/10.3835/plantgenome2016.12.0124
Google Scholar
Pace, J., Yu, X., & Lübberstedt, T. (2015). Genomic prediction of seedling root length in maize (Zea mays L.). The Plant Journal, 83, 903–912. https://doi.org/10.1111/tpj.12937
10.1111/tpj.12937
CASPubMedWeb of Science®Google Scholar
Peiffer, J. A., Flint-Garcia, S. A., De Leon, N., McMullen, M. D., Kaeppler, S. M., & Buckler, E. S. (2013). The genetic architecture of maize stalk strength. PLOS ONE, 8, e67066. https://doi.org/10.1371/journal.pone.0067066
10.1371/journal.pone.0067066
CASPubMedWeb of Science®Google Scholar
Peiffer, J. A., Romay, M. C., Gore, M. A., Flint-Garcia, S. A., Zhang, Z., Millard, M. J., Gardner, C. A. C., McMullen, M. D., Holland, J. B., Bradbury, P. J., & Buckler, E. S. (2014). The genetic architecture of maize height. Genetics, 196, 1337–1356. https://doi.org/10.1534/genetics.113.159152
10.1534/genetics.113.159152
CASPubMedWeb of Science®Google Scholar
Piepho, H.-P., & Möhring, J. (2007). Computing heritability and selection response from unbalanced plant breeding trials. Genetics, 177, 1881–1888. https://doi.org/10.1534/genetics.107.074229
10.1534/genetics.107.074229
PubMedWeb of Science®Google Scholar
Pollak, L. M. (2001). The history and success of the public-private project on germplasm enhancement of maize (GEM). Advances in Agronomy, 78, 45–87. https://doi.org/10.1016/S0065-2113(02)78002-4
Google Scholar
Pugh, N. A., Horne, D. W., Murray, S. C., Carvalho, G. Jr., Malambo, L., Jung, J., Chang, A., Maeda, M., Popescu, S., Chu, T., Starek, M. J., Brewer, M. J., Richardson, G., & Rooney, W. L. (2017). Temporal estimates of crop growth in sorghum and maize breeding enabled by unmanned aerial systems. The Plant Phenome Journal, 1–10. https://doi.org/10.2135/tppj2017.08.0006
Google Scholar
Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M. A. R., Bender, D., Maller, J., Sklar, P., de Bakker, P. I. W., Daly, M. J., & Sham, P. C. (2007). PLINK: A tool set for whole-genome association and population-based linkage analyses. American Journal of Human Genetics, 81, 559–575. https://doi.org/10.1086/519795
10.1086/519795
CASPubMedWeb of Science®Google Scholar
Rincent, R., Charcosset, A., & Moreau, L. (2017). Predicting genomic selection efficiency to optimize calibration set and to assess prediction accuracy in highly structured populations. Theoretical and Applied Genetics, 130, 2231–2247.
10.1007/s00122-017-2956-7
CASPubMedWeb of Science®Google Scholar
Rincent, R., Laloë, D., Nicolas, S., Altmann, T., Brunel, D., Revilla, P., Rodríguez, V. M., Moreno-Gonzalez, J., Melchinger, A., Bauer, E., Schoen, C.-C., Meyer, N., Giauffret, C., Bauland, C., Jamin, P., Laborde, J., Monod, H., Flament, P., Charcosset, A., & Moreau, L. (2012). Maximizing the reliability of genomic selection by optimizing the calibration set of reference individuals: Comparison of methods in two diverse groups of maize inbreds. Genetics, 192, 715–728.
10.1534/genetics.112.141473
CASPubMedWeb of Science®Google Scholar
Romay, M. C., Millard, M. J., Glaubitz, J. C., Peiffer, J. A., Swarts, K. L., Casstevens, T. M., Elshire, R. J., Acharya, C. B., Mitchell, S. E., Flint-Garcia, S. A., McMullen, M. D., Holland, J. B., Buckler, E. S., & Gardner, C. A. (2013). Comprehensive genotyping of the USA national maize inbred seed bank. Genome Biology, 14, R55. https://doi.org/10.1186/gb-2013-14-6-r55
10.1186/gb-2013-14-6-r55
CASPubMedWeb of Science®Google Scholar
Salas Fernandez, M. G., Bao, Y., Tang, L., & Schnable, P. S. (2017). A high-throughput, field-based phenotyping technology for tall biomass crops. Plant Physiology, 174, 2008–2022. https://doi.org/10.1104/pp.17.00707
10.1104/pp.17.00707
PubMedWeb of Science®Google Scholar
Salhuana, W., Sevilla, R., & Eberhart, S. A. (1997). Latin America Maize Project. USDA-ARS.
Google Scholar
Tanksley, S. D., Grandillo, S., Fulton, T. M., Zamir, D., Eshed, Y., Petiard, V., Lopez, J., & Beck-Bunn, T. (1996). Advanced backcross QTL analysis in a cross between an elite processing line of tomato and its wild relative L. pimpinellifolium. Theoretical and Applied Genetics, 92, 213–224. https://doi.org/10.1007/BF00223378
10.1007/BF00223378
CASPubMedWeb of Science®Google Scholar
Tanksley, S. D., & McCouch, S. R. (1997). Seed banks and molecular maps: Unlocking genetic potential from the wild. Science, 277, 1063–1066. https://doi.org/10.1126/science.277.5329.1063
10.1126/science.277.5329.1063
CASPubMedWeb of Science®Google Scholar
Thorwarth, P., Yousef, E. A. A., & Schmid, K. J. (2018). Genomic prediction and association mapping of curd-related traits in gene bank accessions of cauliflower. G3: Genes|Genomes|Genetics, 8, 707–718. https://doi.org/10.1534/g3.117.300199
10.1534/g3.117.300199
Google Scholar
Upadhyaya, H. D., Pundir, R. P. S., Dwivedi, S. L., Gowda, C. L. L., Reddy, V. G., & Singh, S. (2009). Developing a mini core collection of sorghum for diversified utilization of germplasm. Crop Science, 49, 1769–1780. https://doi.org/10.2135/cropsci2009.01.0014
10.2135/cropsci2009.01.0014
Web of Science®Google Scholar
Varsheny, R. K., Bohra, A., Yu, J., Graner, A., Zhang, Q., & Sorrells, M. E. (2021). Designing future 1365 crops: Genomics-assisted breeding comes of age. Trends in Plant Science, 26, 631–649. https://doi.org/10.1111/pbi.13666
Google Scholar
Wang, J., Li, X., Guo, T., Dzievit, M. J., Yu, X., Liu, P., Price, K. P., & Yu, J. (2021). Genetic dissection of seasonal vegetation index dynamics in maize through aerial based high-throughput phenotyping. The Plant Genome, http://doi.org/10.1002/tpg2.20155
Google Scholar
Whittaker, J. C., Thompson, R., & Denham, M. C. (2000). Marker-assisted selection using ridge regression. Genetical Research, 75, 249–252. https://doi.org/10.1017/S0016672399004462
10.1017/S0016672399004462
CASPubMedWeb of Science®Google Scholar
Xiao, J., Li, J., Grandillo, S., Ahn, S. N., Yuan, L., Tanksley, S. D., & McCouch, S. R. (1998). Identification of trait-improving quantitative trait loci alleles from a wild rice relative, Oryza rufipogon. Genetics, 150, 899–909
10.1093/genetics/150.2.899
CASPubMedWeb of Science®Google Scholar
Yu, X., Leiboff, S., Li, X., Guo, T., Ronning, N., Zhang, X., Muehlbauer, G. J., Timmermans, M. C. P., Schnable, P. S., Scanlon, M. J., & Yu, J. (2020). Genomic prediction of maize microphenotypes provides insights for optimizing selection and mining diversity. Plant Biotechnology Journal, 18, 2456–2465. https://doi.org/10.1111/pbi.13420
10.1111/pbi.13420
CASPubMedWeb of Science®Google Scholar
Yu, X., Li, X., Guo, T., Zhu, C., Wu, Y., Mitchell, S. E., Roozeboom, K. L., Wang, D., Wang, M. L., Pederson, G. A., Tesso, T. T., Schnable, P. S., Bernardo, R., & Yu, J. (2016). Genomic prediction contributing to a promising global strategy to turbocharge gene banks. Nature Plants, 2, 16150. https://doi.org/10.1038/nplants.2016.150
10.1038/nplants.2016.150
CASPubMedWeb of Science®Google Scholar
Zhao, K., Tung, C.-W., Eizenga, G. C., Wright, M. H., Ali, M. L., Price, A. H., Norton, G. J., Islam, M. R., Reynolds, A., Mezey, J., McClung, A. M., Bustamante, C. D., & McCouch, S. R. (2011). Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa. Nature Communications, 2, 467. https://doi.org/10.1038/ncomms1467
10.1038/ncomms1467
CASPubMedWeb of Science®Google Scholar

Citing Literature

Volume14, Issue3

November 2021

e20160

This article also appears in:

Advances in Genomic Selection and Application of Machine Learning in Genomic Prediction

Filename	Description
tpg220160-sup-0001-Tables.xlsx1.6 MB	Supplemental Table S1-S5
tpg220160-sup-0002-figuresS1-S12.pdf3.5 MB	Supplemental Figure S1-S12

Journal list menu

Tools

Follow journal

Comprehensive analytical and empirical evaluation of genomic prediction across diverse accessions in maize