WO2016100061A1

WO2016100061A1 - Improved molecular breeding methods

Info

Publication number: WO2016100061A1
Application number: PCT/US2015/064881
Authority: WO
Inventors: Frank TECHNOW; Liviu Radu TOTIR
Original assignee: Pioneer Hi Bred International Inc
Current assignee: Pioneer Hi Bred International Inc
Priority date: 2014-12-18
Filing date: 2015-12-10
Publication date: 2016-06-23
Anticipated expiration: 2017-06-18
Also published as: CL2017001538A1; AU2015362942B2; CA2968120A1; BR112017012891B1; AR103075A1; US11980147B2; CN107205352A; MX2017007712A; BR112017012891A2; AU2015362942A1; US20170359978A1

Abstract

A Bayesian multilevel whole-genome regression model is disclosed and its prediction performance compared to that of the popular BayesA model applied to each population separately (no pooling) and to the joined data set (complete pooling). For small population sizes (e.g., < 50), partial pooling increased prediction accuracy over no or complete pooling for populations represented in the estimation set. Partial pooling with multilevel models can make optimal use of information in multi-population estimation sets.

Description

IMPROVED MOLECULAR BREEDING METHODS

Background

[1] Estimation set size is an important determinant of genomic prediction accuracy. Plant breeding programs are characterized by a high degree of structuring, particularly into populations. This hampers establishment of large estimation sets for each population.

[2] The use of genomic selection in animal and plant breeding is based on the ability to generate accurate genomic estimated breeding values (GEBV). An important determinant of prediction accuracy is the size of the estimation set. In animal breeding, assembling large single-breed estimation sets is relatively straight forward for dairy breeds like Holstein Friesian, where genomic selection is applied most successfully to date. However, for some dairy cattle breeds and in particular for beef cattle breeds, assembling single-breed estimation sets of sufficient size is often not possible. Creation of multi-breed estimation sets by pooling data from several breeds together, is therefore of great interest and subject of active research.

[3] A similar situation exists in plant breeding, which is characterized by a high degree of breeder induced structure. This structure results from the importance of keeping distinct heterotic groups for maximum exploitation of heterosis, from use of distinct biparental breeding populations and the need for specialized breeding programs targeting specific traits or environments. This requires that the phenotyping and genotyping resources available to a breeding program have to be allocated to multiple populations, which can prevent the creation of large estimation sets for each population. Several studies therefore investigated the merit of pooled estimation sets combining populations or even heterotic groups. [4] However, pooling estimation sets is complicated by differences in genetic properties among populations, such as pertaining to linkage disequilibrium, allele frequencies or relationship structure. This might be the reason that using pooled estimation sets failed to increase prediction accuracy in some applications in plant and animal breeding.

Description of the Figures

[5] FIG. 1 is a graphical visualization of the multilevel model (A) and the conventional BayesA model (B).

[6] FIG. 2 is a graphical visualization of the testing strategy for evaluating prediction accuracy. The estimation set comprises A, and A₂ from populations Λ and P₂ (set Π). The prediction accuracy of lines from populations represented in estimation set (r_n) was computed from A, and A₂, the prediction accuracy of lines from populations not represented in estimation set from lines in P, and P₄ (set Π).

Detailed Description

[7] It has been proposed to use separate estimation sets for each population but to derive genome position specific priors using data from other populations. In this way, unique genome properties of each population could be accounted for while still using information from other populations. A similar, more formal approach is partial pooling, facilitated by Bayesian multilevel models. In multilevel models, specific marker effects are estimated for each population. However, the prior means of these specific marker effects, which might be interpreted as overall or unspecific marker effects, are estimated from data of all populations, simultaneously with the unspecific marker effects. Because the specific marker effects are shrunk towards the overall effects, the former are still informed by data from the other populations to a certain degree. Partial pooling thus strikes a middle ground between no pooling (specific marker effects estimated from data of the specific population only) and complete pooling (common marker effects estimated from pooled estimation sets).

[8] Pooling populations increases estimation set size but ignores unique genetic characteristics of each. A possible solution is partial pooling with multilevel models, which allows estimating population specific marker effects while still leveraging information across populations.

[9] The objectives were to (i) demonstrate the use of Bayesian multilevel whole- genome regression models for genomic prediction and (ii) investigate scenarios in which partial pooling might be superior over no or complete pooling of estimation sets. These investigations were based on two publicly available maize breeding data sets and supported by a simulation study.

Materials and Methods

Multilevel whole genome regression model

[10] The statistical model fitted to the data is y, ~ N(j ,„o_e ²)

I¹'! = βο + ∑k ^zijk^ujk

where y,, is the observed phenotypic value of the i'^h individual from the j'^h population and μ,, its linear predictor. The phenotypic data y, was centered to mean zero and scaled to unit variance. The Normal density function, which is used as data model, is denoted as N with σ denoting the residual variance. The common intercept was β₀. Finally, u_lt denotes the additive effect of the k^h biallelic single nucleotide polymorphism (SNP) marker in population j. The genotype of individual i from population j at marker k was represented by z_≠ and it denotes the number of reference alleles, centered by twice the reference allele frequency. Which of the alleles was chosen as reference allele depended on the data set and is described below. Effects u_lt were only estimated when the corresponding marker k was polymorphic in population j. Otherwise it was set to 0 and treated as a constant.

[11] A graphical display of the hierarchical prior distribution setup is shown in Figure 1 A. The prior of u_lt is

Uj_k ~ N(u_k, Y² _k) ^ where u_t is the overall effect of the k^h marker and the variance parameter y ί quantifies the deviations of the specific effects u_lk from u_t. The shrinkage toward u_t is the stronger the smaller y². Both parameters, u _t and y are associated with prior distributions themselves and estimated from the data. For u_t this is u_t ~ N(0, σ .). Here, the variance parameter a_t ² controls the amount of shrinkage towards 0. It is associated with a scaled inverse Chi-square prior with 4.001 degree of freedom and scale parameter ². The prior for u _t thus corresponds to the "BayesA" prior.

[12] For the variance parameter y ,

Y_k ~ N(m, d², 0 < a, b =∞)

·*) which is a Normal distribution prior on y_t with mean parameter m and standard deviation d, left truncated at zero. The mean of the truncated distribution N(m, d², 0 < a, b =∞), which is a function of m, d and the truncation points a and b, can be interpreted as the "typical" deviation of the specific marker effects u_lt from u_t. Higher values of the mean of N(m, d², 0 < a,b = co) indicate larger deviations and vice versa. This parameter might therefore be used to quantify population divergence.

[13] An uniform prior Uni(0.00\, 0.5) is used for the hyperparameters S², m and d. The prior for the intercept β₀ is a Normal distribution with mean 0 and a very large variance. For the residual variance σ , a uniform distribution prior over the interval [0, 1] on o_e was specified, which agrees with recommendations for uninformative priors on variance components.

[14] Samples from the posterior distribution were drawn with Gibbs sampling, implemented in the JAGS Gibbs sampling environment. The total number of samples used for inference was 1000, drawn from a single chain of length 510000. The first 10000 samples were discarded as burn in and only every 500th subsequent sample stored. These settings ensured convergence and an effective sample size (ESS) of > 100 for all parameters (ESS of u_t and u_lt were typically > 500).

[15] The ESS was calculated with the R package CODA, which was also used to monitor convergence using diagnostic plots.

Conventional whole genome regression model

[16] The Bayesian whole genome regression method "BayesA" was used, with modifications pertaining to the hyperparameter S¹ (see Figure IB for a graphical representation). The linear model is

y„ ~ N(„a² _e) ^ μ,, = β« +∑k ^zijk^uk

which is the same as model (1), except that the population index j is dropped from the marker effects u_t. For no pooling, the model was applied to each population in turn, for complete pooling to the joint data set. For σ₈ ² an improper scaled inverse Chi-square prior with -1 degrees of freedom and scale equal to zero was used. This is equivalent to a uniform prior on σ„, as is used for the multilevel model, but exploits conjugancy. [17] The BayesA Gibbs Sampler was implemented as a C routine compatible with the R statistical software environment. Again, the total number of samples used for inference was 1000, drawn from a single chain of length 510000, with burn in of 10000 and thinning intervals of 500.

Estimation, prediction and testing procedure

[18] Let Π denote the set of P populations represented in the estimation set and the set of N_P individuals from a population in Π as Λ_Ρ, where p indexes the populations in Π. A graphical representation is presented in Figure 2. Further, let those individuals from a population in Π that are not in Λ_Ρ be denoted as Λ_Ρ and the set of populations not in Π as Π. Populations in Π will be referred to as "new" populations. The estimation set thus comprised all individuals belonging to Λ_Ρ , for p G Π. The test set used for calculating prediction accuracy, comprised individuals in Λ_Ρ from populations in Π and all individuals from populations in Π. The phenotypic observations of test individuals were masked in the estimation procedure. The separation of populations into Π and Π and of individuals within a population into Λ_Ρ and Λ_Ρ was done at random.

[19] Within each population, prediction accuracy was computed as the correlation between GEBVs and observed phenotypic values of individuals in the testing set. The within population prediction accuracies were subsequently averaged for populations in Π and Π. These average within population prediction accuracies will henceforth be denoted as r_n and r^. Thus, r_n and correspond to the prediction accuracy for populations represented and not represented in the estimation set, respectively.

[20] When using partial pooling, GEBVs of individuals in Λ_Ρ were predicted using the posterior means of the marker effects estimated for the corresponding population {i.e., u_Jt). GEBVs of individuals from populations in Π were predicted using the posterior means of the overall (unspecific) marker effects u_t.

[21] When using complete pooling, GEBVs of all individuals in the test set were predicted from the posterior means of marker effects u_t estimated from the joint data set with model (4).

[22] Finally, when using no pooling, GEBVs of individuals in Λ_Ρ were predicted using the posterior means of the marker effects u_t obtained after applying model (4) to the estimation data from the corresponding set Λ_Ρ. The no pooling approach does not provide a direct way of predicting GEBVs of individuals from populations in Π. Thus, was not evaluated for the no pooling approach.

Application to nested association mapping (NAM) maize populations

[23] The NAM data set was obtained from http : //www.panzea . org. It comprised 4699 recombinant inbred lines (RILs) from 25 biparental crosses between a genetically diverse set of maize inbred lines and line B73 as common parent. The average population size was 188. The RILs were genotyped with 1106 polymorphic SNP markers covering the whole genome. The non-B73 allele was defined as the reference allele. All SNP were biallelic and thereby the reference allele corresponded to the same nucleotide in all 25 populations. To facilitate computations, a thinned set of 285 markers was used, chosen in such a way that there was one marker per 5 cM interval, on average. A density of one marker per 10 cM interval is sufficient for genomic prediction in the NAM population. The traits days to silking (DS), ear height (EH), ear length (EL), southern leaf blight resistance (SLB), near-infrared starch measurements (NS) and upper leaf angle (ULA) were analyzed and phenotyped in multi-environment field trials. The phenotypic records used for fitting the models were averages over the single environment phenotypes. The number of environments were 10, 11, 8, 3, 7 and 9 for DS, EH, EL, SLB, NS and ULA, respectively. The traits chosen represent the major trait categories available: yield component (EL), agronomic (EH), disease resistance (SLB), flowering (DS), quality (NS) and morphology (ULA).

[24] To investigate the effect of total number of lines N, number of populations P and number of lines per population N_p in the estimation set on prediction accuracy and the relative performance of the pooling approaches, the following combinations of P and N_p were considered: P = 5 and N_P = 50 and 100, P = 10 and N_P = 25, 50 and 100, P = 20 and N_P = 12.5, 25, and 50. For P = 20 and N_p = 12.5, we sampled 19 populations with 12 individuals and one with 22, which results in an average N_p of 12.5. The P and N_p combinations thus gave rise to N of either 250, 500 or 1000. For each combination of trait, P and N_p, 50 estimation-testing data sets were generated by repeating the sampling of Π and Λ_Ρ as described above. Throughout, the three pooling approaches were applied to the same data sets. The sampling variation between different data sets thus does not enter the comparisons among pooling approaches. Application to interconnected biparental (IB) maize populations

[25] This data set was obtained from the supplement of Riedelsheimer et al. (2013). It comprised 635 doubled haploid (DH) lines from five biparental populations with average size of 127. The populations were derived from crosses between four European flint inbred lines. For all DH lines 16741 SNP markers polymorphic across populations were available. Missing marker genotypes were replaced with twice the frequency of the reference allele, which was the allele with the lower frequency. When analyzing the data we used a thinned set of 285 markers. Because the data set did not include a map of the markers, the markers were chosen randomly.

[26] The DH lines were phenotyped in multi-environment field trials for Giberella ear rot severity (GER), a fungal disease caused by Fusarium graminearum, deoxynivalenol content (DON, a major mycotoxin produced by the fungus), ear length (EL), kernel rows (KR) and kernels per row (KpR). A more detailed description of this data set can be found in Riedelsheimer et al. (2013) and Martin et al. (2012).

[27] As described above, populations were randomly split into Λ_Ρ and Λ_Ρ. However, because there were only five populations in total, no populations from Π were excluded. Set Π was thus empty and was not evaluated.

[28] The sets Λ_Ρ comprised 25%, 50% and 75% of the lines in each population, which corresponded to an average N_p of 31, 63 and 95, respectively. For each trait and percentage value of estimation individuals, 100 estimation-testing data sets generated, each time resampling the subset of 285 markers too.

Application to simulated data set

[29] A simulation study was conducted to specifically investigate the performance of the pooling approaches under increasing levels of differences in QTL effects among populations. The bases for the simulation were the marker genotypes of the lines in the NAM populations. To simulate genetic values, 20 marker loci were randomly chosen as QTL, which were subsequently removed from the set of observed markers. Additive overall effects a_q were drawn from a standard normal distribution. Then population specific QTL effects a_iq were sampled from N(a_q, T_q ²). The variance parameter τ² was chosen such that the relative standard deviation (rSD), i.e., x_q/a_q , was equal to 2, 1, 0.5, 0.25 and 0.0. The greater rSD, the less similar the population specific QTL effects are. True genetic values were obtained by summing QTL effects a_iq according to the QTL genotypes of each individual. Finally phenotypic values were simulated by adding a normally distributed noise variable to the true genetic values. The variance of the noise variable was chosen such that the heritability across populations was equal to 0.70. The average within family heritability necessarily increased with decreasing rSD, and was 0.53, 0.58, 0.64, 0.68 and 0.70 at rSD 2, 1, 0.5, 0.25 and 0.0, respectively.

[30] Set Π comprised P = 10 populations and sets Λ_Ρ had size N_p = 25. For each rSD value 50 estimation-testing data sets were generated. The QTL positions and effects were randomly generated anew for each data set. Also in this case a thinned set of 285 markers was used. Because the true genetic values were known, r_n and r^ were computed as the correlation between true genetic values and GEBVs.

Results

NAM maize populations

[31] Trends typically held across traits. The results presented and discussed therefore apply to all traits, unless otherwise mentioned.

[32] Increasing N_p while keeping N constant (i.e., having fewer but larger populations in the estimation set) generally increased , r_n and decreased , (Table I). However, the increase in r_n was much more pronounced than the decrease in ¾.

[33] When increasing N_p with constant P or when increasing P with constant N_p, both r_n and increased (Table V). However, while in the first case, r_n and increased in similar magnitudes, the increase in r_n was much smaller than the increase in in the second case, in particular when N_p was high. Per definition, the accuracy of no pooling is not expected to change as long as N_p remains constant.

[34] For low P and high N_p, e.g., P = 5 and N_p = 100, no pooling achieved the highest r_n and complete pooling the lowest (Table V). For high P and low N_p, e.g., P = 20 and N_p = 25, partial pooling achieved the highest r_n . Here no pooling resulted in the lowest r_n. The only exception to this was trait DS, where no pooling had a r_n equal or higher to partial and complete pooling also for low N_p. [35] Partial and complete pooling achieved virtually identical prediction accuracies 7 fj- for new populations (Table 1). In general, of a particular pooling approach was considerably lower than the corresponding r_n. The differences between r_n and tended to be larger for high N_p.

Table 1: Average within population prediction accuracies in NAM maize populations. Values shown are average within population prediction accuracies for test individuals, averaged over 50 random estimation-test data splits. The standard errors were < 0.013. P gives the size of set Π, i.e., the number of populations represented in the estimation set, column Np gives the number of individuals from each population in Π that were used for estimation, i.e., the sizes of sets Λρ. The traits were: days to silking (DS), ear height (EH), ear length (EL), southern leaf blight resistance (SLB), near-infrared starch measurements (NS) and upper leaf angle (ULA).

P trait no partial complete partial comp

5 50 DS 0.41 0.34 0.26 0.19 0.19

EH 0.47 0.44 0.39 0.31 0.32

EL 0.39 0.37 0.28 0.19 0.19

NS 0.39 0.37 0.32 0.25 0.26

SLB 0.49 0.49 0.45 0.37 0.37

ULA 0.50 0.48 0.44 0.36 0.36

100 DS 0.52 0.41 0.28 0.21 0.20

EH 0.57 0.51 0.43 0.34 0.34

EL 0.49 0.46 0.35 0.23 0.23

NS 0.47 0.44 0.36 0.29 0.29

SLB 0.58 0.58 0.50 0.41 0.41

ULA 0.58 0.54 0.47 0.40 0.40

10 25 DS 0.32 0.28 0.22 0.18 0.17

EH 0.38 0.38 0.35 0.30 0.31

EL 0.31 0.31 0.25 0.21 0.21

NS 0.30 0.33 0.30 0.26 0.27

SLB 0.40 0.46 0.43 0.38 0.39

ULA 0.39 0.44 0.41 0.36 0.37

50 DS 0.42 0.35 0.26 0.22 0.22

EH 0.47 0.45 0.40 0.36 0.36

EL 0.40 0.39 0.29 0.23 0.23

NS 0.38 0.40 0.35 0.30 0.30

SLB 0.49 0.52 0.46 0.42 0.43

ULA 0.48 0.50 0.45 0.41 0.41 100 DS 0.51 0.42 0.30 0.25 0.25

EH 0.57 0.53 0.44 0.39 0.39

EL 0.48 0.46 0.33 0.27 0.27

NS 0.48 0.46 0.38 0.33 0.33

SLB 0.57 0.57 0.49 0.45 0.45

ULA 0.59 0.57 0.48 0.44 0.44

20 12.5 DS 0.23 0.23 0.21 0.17 0.17

EH 0.28 0.34 0.33 0.30 0.31

EL 0.22 0.27 0.23 0.19 0.19

NS 0.21 0.30 0.29 0.27 0.28

SLB 0.31 0.43 0.42 0.38 0.39

ULA 0.28 0.40 0.39 0.35 0.36

25 DS 0.32 0.30 0.24 0.22 0.23

EH 0.38 0.42 0.39 0.36 0.37

EL 0.31 0.34 0.28 0.22 0.22

NS 0.30 0.36 0.33 0.30 0.31

SLB 0.39 0.48 0.45 0.42 0.43

ULA 0.38 0.46 0.44 0.42 0.42

50 DS 0.42 0.37 0.29 0.26 0.26

EH 0.48 0.49 0.42 0.40 0.40

EL 0.39 0.40 0.30 0.28 0.29

NS 0.38 0.41 0.36 0.34 0.34

SLB 0.49 0.54 0.48 0.46 0.47

ULA 0.49 0.52 0.48 0.46 0.46

IB maize populations

[36] The prediction accuracy r_n increased with increasing N_p, for all traits and pooling approaches (Table 2). Averaged over traits, the increase was largest for no pooling, where the accuracy increased from an average of 0.35 at N_p = 31 to 0.48 at N_p = 95. The accuracies for the partial and complete pooling approaches increased from 0.39 and 0.38, respectively, at N_P = 31 to 0.48 at N_P = 95.

[37] At Np = 31, partial pooling had the highest r_n for traits EL, KpR, complete pooling for traits DON and KR. For GER both had the same accuracy. The no pooling approach had the lowest r_n, except for EL and KpR, where it had the same accuracy as complete pooling. For the highest N_p of 95, the accuracy differences among the pooling approaches decreased. Partial pooling still had the highest accuracy for EL and KpR and the same as complete pooling for DON and GER. While never better than partial pooling, no pooling had higher prediction accuracy than complete pooling for EL and KpR.

Table 2: Average within population prediction accuracies in interconnected biparental maize populations. Values shown are average within population prediction accuracies for test individuals, averaged over 100 random estimation-test data splits. Standard errors were < 0.01. N_p denotes the average number of individuals per population in the estimation set. The traits were ear length (EL), deoxynivalenol content (DON), Giberella ear rot severity (GER) kernel rows (KR) and kernels per row (KpR).

K Trait Pooling

no partial complete

31 EL 0.31 0.33 0.31

DON 0.38 0.44 0.46

GER 0.38 0.43 0.43

KR 0.46 0.50 0.52

KpR 0.21 0.23 0.21

62 EL 0.40 0.41 0.39

DON 0.47 0.51 0.51

GER 0.47 0.50 0.49

KR 0.53 0.56 0.58

KpR 0.28 0.29 0.27

95 EL 0.44 0.46 0.43

DON 0.51 0.53 0.53

GER 0.51 0.53 0.53

KR 0.56 0.58 0.59

KpR 0.31 0.32 0.30

Simulated maize populations

[38] For all pooling approaches, r_n increased with decreasing rSD (Table 3). The increase for no pooling, however, was comparatively small and a result of the increasing within family heritability with decreasing rSD. The relative performance of the pooling approaches also depended on rSD. For the highest rSD value considered, no pooling had the highest r_n, for the intermediate rSD value of 1.0 partial pooling. For the lower rSD values complete and partial pooling achieved similarly high r_n . [39] Also for both partial and complete pooling increased strongly with decreasing rSD and the differences to r_n decreased (Table 3). Partial and complete pooling achieved almost identical r^.

[40] The mean of the truncated Normal distribution prior N(m, d², 0 < a,b =∞) for parameter y_k increased with increasing rSD. Its average values were 0.0111, 0.0153, 0.0190, 0.0269 and 0.0296 for rSD of 0.0, 0.25, 0.5, 1.0 and 2.0, respectively.

Table 3: Average prediction accuracies for simulated maize populations. Values shown are average within population prediction accuracies for test individuals, averaged over 50 random estimation-test data splits. Standard errors were < 0.015. rSD is the relative standard deviation of simulated population specific QTL effects. rSD no partial complete partial complete

0.0 0.54 0.89 0.89 0.89 0.89

0.25 0.51 0.84 0.85 0.84 0.84

0.5 0.50 0.76 0.76 0.73 0.73

1.0 0.48 0.57 0.53 0.48 0.49

2.0 0.44 0.41 0.30 0.20 0.21

Discussion

Comparison of pooling approaches

[41] Partial pooling allows estimation of population specific marker effects while still facilitating "borrowing" of information across populations. It is therefore a compromise between no pooling, which models unique characteristics of each population but ignores shared information, and complete pooling, in which the opposite is the case.

[42] When population sizes N_p are sufficiently large, borrowing information from other populations is not required for achieving high prediction accuracy of new individuals from the same population (r_n). Further enlarging estimation sets by pooling with other populations might then even be detrimental. This explains why no pooling was the most accurate approach when N_p was large (e.g., >= 50), particularly in the NAM population, and why it profited most from increases in N_p. Therefore, pooling of estimation sets is most promising if N_p is small due to budget or other constraints. It was observed that pooling was more accurate than no pooling when N_p was small (e.g., < 50). The superiority of either pooling approach over no pooling also increased with increasing P , because information from more populations was available, which is not used in no pooling. Thus, pooling is expected to most advantageous when P is relatively high and N_p low. Whether partial or complete pooling is the better approach will then also depend on the similarity of the pooled populations. The greater the similarity, the relatively better complete pooling is expected to perform, because the ability to estimate population specific marker effects becomes less important. In this situation partial pooling might even be of disadvantage, because it requires estimation of many more effects which might lead to problems associated with nonidentifiability. The parents of the IB populations come from the same breeding program, whereas the non-common parents of the NAM populations were chosen to be maximally diverse and comprise temperate, tropical and specialty (sweet and popcorn) maize germplasm. Accommodating for unique characteristics of the populations is therefore more important in NAM than in IB, which might explain why complete pooling was always inferior to partial pooling in the former but often equal or even superior in the latter and also why no pooling never achieved the highest prediction accuracy in IB, even for large N_p.

[43] The relative performance of the pooling approaches was very stable across traits in the NAM data set, with the exception of DS. For this trait the no pooling approach was generally superior, even at high P and low N_p. Buckler et al. (2009) found evidence for an allelic series at the QTL identified for DS in the NAM population. Thus, while the positions of the QTL are conserved across populations, their effects differ strongly. Possible reasons are presence of multiple alleles or QTL by genetic background interaction. In this situation, pooling of data is not expected to have an advantage over no pooling. This example also shows that decisions about whether to pool data or not have to be made on a by trait basis and should incorporate prior knowledge about genetic architecture, if available.

[44] The dependence of the relative performance of the pooling approaches on the similarity of genetic effects among populations was also reinforced by the results from the simulation study. There it was also observed that the mean of N(m, d², 0 < a,b =∞), the prior distribution of

which quantifies the deviations of specific marker effects u_lk from the overall effect u_k, increased with increasing simulated differences among population specific QTL effects. This was expected, but demonstrates that the data was informative for the high level hyperparameters. Averaged over P and N_p, this mean was largest for DS and ULA in NAM (results not shown). This might reflect the noted differences between population specific QTL effects for DS. Trait ULA, however, did not diverge from the prediction accuracy pattern observed for the remainder of traits and there does not seem to be any strong indication of an allelic series as in DS. There was also no obvious relation between the mean of N(m, d², 0 < a,b =∞) and performance of the pooling approaches in IB (results not shown).

[45] Modeling unique characteristics of populations requires that these populations are represented in the estimation set. Prediction of individuals from new populations in Π therefore has to rely on the overall, unspecific marker effects u_k, in both partial and complete pooling. It was thus expected that both achieved very similar prediction accuracies for new populations.

[46] These results demonstrate that partial pooling is able to model unique characteristics of populations within the estimation set without compromising on the ability of prediction of individuals from new populations.

[47] This study examplified the use of multilevel models for partial pooling in the context of multiple populations, a scenario of high relevance for plant and animal breeding. However, the concept is readily applicable in a wide array of scenarios. Examples are pooling data across multiple top-cross testers or environments, as is of particular relevance in plant breeding. Extending the models to more than two levels is straightforward, too, for example for pooling multiple populations from multiple heterotic groups or breeding programs.

Composition of estimation set

[48] Increasing the number of individuals from a population in the estimation set (N_p) always increased prediction accuracy for untested individuals from the same population (r_n), regardless if the estimation set was further enlarged by individuals from other populations (partial and complete pooling) or not (no pooling).

[49] However, because plant breeding programs have to operate under budget constraints, optimum allocation of resources is of great importance for maximizing the potential of genomic selection. With a fixed budget for phenotying that is proportional to N, the number of populations P and the number of individuals per population N_p have to be optimized under the constraint that N = P^■ N_p. Such an optimization could be accomplished using basic theory about response to selection and accounting for the different prediction accuracy for populations represented and not represented in the estimation set (r_n and r^, respectively). A key point hereby is that r_n will increase with increasing N_p but it will apply to fewer populations because of the decrease in P. This is exacerbated by the decrease in that we observed was associated with decreasing P. Thus, if the total number of populations is large, as is typically the case in plant breeding programs, having very low P is likely to be undesirable. In the context of plant breeding this study showed that pooling data across populations can at least partly compensate for low N_p if populations are related and there is evidence for the merit of pooling very divergent germplasm too. Using pooled estimation sets therefore has the potential to allow for high P without compromising too much on r_n. This study showed that partial pooling with multilevel models can further enhance this potential by making optimal use of the information in pooled estimation sets.

Claims

CLAIMS What is claimed is:

1. A method for selecting individuals in a breeding program and for breeding selected individuals, said method comprising: a. constructing an optimized estimation data set for whole genome prediction and respectively selection by: i. selecting and pooling predefined candidate sets of populations, said predefined candidate sets being selected from breeding crosses, and selecting individuals within each breeding cross for phenotyping from a comprehensive list of individuals targeted for selection wherein genotypic information is available for all candidate individuals; ii. computing, for a population of breeding crosses and breeding traits under active artificial selection, specific genetic similarity or divergence measures using objective functions of population and trait specific estimated QTL effects; iii. constructing optimized estimation datasets in terms of number of populations pooled and respective individuals within each population; and iv. using no pooling, partial pooling, or complete pooling statistical models as a function of objective genetic similarity or divergence criteria; b. phenotyping candidates in the optimized estimation data set; c. genotyping breeding individuals at a plurality of markers; d. obtaining genomic estimated breeding values for the genotyped breeding individuals using the phenotypes of the candidates in the optimized estimation data sets;

e. making selections within the complete set of genotyped candidate for selection individuals based on the genomic estimated breeding values generated conditional on optimized estimation sets through no, partial or complete pooling; and

f. breeding at least one of the selection individuals with at least one other of the selection individuals.

2. The method of claim 1, wherein said genotypic information for the candidate is obtained via genotyping using S P markers.

3. The method of claim 1, wherein said breeding individuals are homozygous.

4. The method of claim 1, wherein said plant is selected from the group consisting of: maize, soybean, sunflower, sorghum, canola, wheat, alfalfa, cotton, rice, barley, millet, sugar cane and switchgrass.

5. The method of claim 1, wherein said breeding individuals are animals.

6. The method of claim 1, wherein the method is applied to plant breeding.

7. The method of claim 1, wherein the method is applied to animal breeding.

8. The method of claim 1, further comprising a genetically diverse population that includes individuals carrying one or more transgenes.

9. The method of claim 1, further comprising a genetically diverse population that includes individuals with DNA edited with Cas9.

10. The method of claim 1, wherein said genotypic information for the candidate is obtained by analyses of gene expression, metabolite concentration, or protein

concentration.