Using PopVar

Introduction

To make progress in breeding, populations should have a favorable mean and high genetic variance (Bernardo 2010). These two parameters can be combined into a single measure called the usefulness criterion (Schnell and Utz 1975), visualized in Figure 1.

Figure 1. Visualization of the mean, genetic variance, and superior progeny mean of a single population.

Ideally, breeders would identify the set of parent combinations that, when realized in a cross, would give rise to populations meeting these requirements. PopVar is a package that uses phenotypic and genomewide marker data on a set of candidate parents to predict the mean, genetic variance, and superior progeny mean in bi-parental or multi-parental populations. Thre package also contains functionality for performing cross-validation to determine the suitability of different statistical models. More details are available in Mohammadi, Tiede, and Smith (2015). A dataset think_barley is included for reference and examples.

Installation

You can install the released version of PopVar from CRAN with:

install.packages("PopVar")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("UMN-BarleyOatSilphium/PopVar")

Functions

Below is a description of the functions provided in PopVar:

Function Description
pop.predict Uses simulations to make predictions in recombinant inbred line populations; can internally perform cross-validation for model selections; can be quite slow.
pop.predict2 Uses deterministic equations to make predictions in populations of complete or partial selfing and with or without the induction of doubled haploids; is much faster than pop.predict; does not perform cross-validation or model selection internally.
pop_predict2 Has the same functionality as pop.predict2, but accepts genomewide marker data in a simpler matrix format.
x.val Performs cross-validation to estimate model performance.
mppop.predict Uses deterministic equations to make predictions in 2- or 4-way populations of complete or partial selfing and with or without the induction of doubled haploids; does not perform cross-validation or model selection internally.
mpop_predict2 Has the same functionality as mppop.predict, but accepts genomewide marker data in a simpler matrix format.

Examples

Below are some example uses of the functions in PopVar:

# Load the package
library(PopVar)

# Load the example data
data("think_barley", package = "PopVar")

Predictions using simulated populations

The code below simulates a single population of 1000 individuals for each of 150 crosses. For the sake of speed, the marker effects are predicted using RR-BLUP and no cross-validation is performed.

out <- pop.predict(G.in = G.in_ex, y.in = y.in_ex, map.in = map.in_ex,
                   crossing.table = cross.tab_ex,
                   nInd = 1000, nSim = 1, 
                   nCV.iter = 1, models = "rrBLUP")

The function returns a list, one element of which is called predictions. This element is itself a list of matrices containing the predictions for each trait. They can be combined as such:

predictions1 <- lapply(X = out$predictions, FUN = function(x) {
  x1 <- as.data.frame(apply(X = x, MARGIN = 2, FUN = unlist), stringsAsFactors = FALSE)
  cbind(x1[,c("Par1", "Par2")], sapply(X = x1[,-1:-2], as.numeric)) 
})

# Display the first few lines of the predictions for grain yield
knitr::kable(head(predictions1$Yield_param.df))
Par1 Par2 midPar.Pheno midPar.GEBV pred.mu pred.mu_sd pred.varG pred.varG_sd mu.sp_low mu.sp_high low.resp_FHB low.resp_DON low.resp_Height high.resp_FHB high.resp_DON high.resp_Height cor_w/_FHB cor_w/_DON cor_w/_Height
FEG66-08 MN97-125 99.200 100.65607 100.58065 NA 7.564859 NA 95.81135 105.2983 23.16470 23.84100 78.78625 25.01624 25.13016 75.91027 0.2572561 0.2422184 -0.3199704
MN99-71 FEG90-31 NaN 99.17635 99.12126 NA 5.255394 NA 95.36126 103.0920 21.57383 23.55555 77.93688 24.66107 25.09007 75.62009 0.4670046 0.1604766 -0.1805932
MN96-141 FEG183-52 NaN 101.28761 101.24542 NA 5.664738 NA 97.12738 105.3532 23.74467 23.78425 79.70719 27.32846 28.44027 76.12451 0.7288072 0.7473355 -0.5407719
MN99-02 FEG183-52 NaN 100.78451 100.71886 NA 10.085053 NA 95.33375 106.1531 25.71896 23.94873 74.93540 26.41312 26.65181 73.33489 0.1153523 0.3903699 -0.1438686
FEG99-10 FEG148-56 NaN 98.68505 98.78295 NA 4.046123 NA 95.36134 102.2904 19.56935 20.26895 85.53316 23.07361 24.19603 80.09071 0.5620461 0.6152524 -0.5809222
MN99-62 MN01-46 105.775 102.29724 102.27109 NA 5.045291 NA 98.53626 106.0721 26.09207 28.39819 73.20279 26.39339 28.49207 74.12016 0.2088950 -0.1179521 0.3326492

Predictions using deterministic equations

Generating predictions via simulated populations can become computationally burdensome when many thousands or hundreds of thousands of crosses are possible. Fortunately, deterministic equations are available to generate equivalent predictions in a fraction of the time. These equations are provided in the pop.predict2 and pop_predict2 functions.

The pop.predict2 function takes arguments in the same format as pop.predict. We have eliminated the arguments for marker filtering and imputation and cross-validation, as the pop.predict2 function does not support these steps. (You may continue to conduct cross-validation using the x.val function.) Therefore, the genotype data input for pop.predict2 must not contain any missing data. Further, these predictions assume fully inbred parents, so marker genotypes must only be coded as -1 or 1. The data G.in_ex_imputed contains genotype data that is formatted properly.

Below is an example of using the pop.predict2 function:

out2 <- pop.predict2(G.in = G.in_ex_imputed, y.in = y.in_ex, map.in = map.in_ex,
                     crossing.table = cross.tab_ex, models = "rrBLUP")

knitr::kable(head(subset(out2, trait == "Yield")))
parent1 parent2 trait pred_mu pred_varG pred_musp_low pred_musp_high cor_w_FHB cor_w_DON cor_w_Yield cor_w_Height pred_cor_musp_low_FHB pred_cor_musp_low_DON pred_cor_musp_low_Yield pred_cor_musp_low_Height pred_cor_musp_high_FHB pred_cor_musp_high_DON pred_cor_musp_high_Yield pred_cor_musp_high_Height
3 FEG66-08 MN97-125 Yield 100.41166 7.768361 95.52021 105.3031 0.3412831 0.3303378 NA -0.4004154 22.27035 23.76433 NA 78.84012 24.96750 25.91641 NA 75.11838
7 MN99-71 FEG90-31 Yield 98.98546 5.298104 94.94591 103.0250 0.4959745 0.2776666 NA -0.1612311 21.86563 23.41697 NA 77.04669 24.85938 25.18159 NA 75.75153
11 MN96-141 FEG183-52 Yield 101.07137 5.571654 96.92884 105.2139 0.6132612 0.6961326 NA -0.4603399 24.36629 23.67733 NA 79.59345 27.88028 28.40806 NA 75.16009
15 MN99-02 FEG183-52 Yield 100.82967 9.499736 95.42053 106.2388 0.0249962 0.4211875 NA -0.1929275 26.15174 23.73807 NA 74.76672 26.29180 26.60344 NA 72.86270
19 FEG99-10 FEG148-56 Yield 98.01593 4.121802 94.45292 101.5789 0.5959335 0.6401323 NA -0.5080571 19.67552 19.58928 NA 85.81854 23.38585 24.23917 NA 80.69997
23 MN99-62 MN01-46 Yield 102.51094 4.673196 98.71709 106.3048 0.0755600 -0.0827658 NA 0.4010907 26.07342 28.37124 NA 72.67125 26.20580 28.16102 NA 74.51340

Note that the output of pop.predict2 is no longer a list, but a data frame containing the combined predictions for all traits.

The formatting requirements of G.in for pop.predict and pop.predict2 are admittedly confusing. Marker genotype data is commonly stored as a n x p matrix, where n is the number of entries and p the number of markers. The function pop_predict2 accommodates this general marker data storage. Here is an example:

out3 <- pop_predict2(M = G.in_ex_mat, y.in = y.in_ex, map.in = map.in_ex,
                     crossing.table = cross.tab_ex, models = "rrBLUP")

knitr::kable(head(subset(out2, trait == "Yield")))
parent1 parent2 trait pred_mu pred_varG pred_musp_low pred_musp_high cor_w_FHB cor_w_DON cor_w_Yield cor_w_Height pred_cor_musp_low_FHB pred_cor_musp_low_DON pred_cor_musp_low_Yield pred_cor_musp_low_Height pred_cor_musp_high_FHB pred_cor_musp_high_DON pred_cor_musp_high_Yield pred_cor_musp_high_Height
3 FEG66-08 MN97-125 Yield 100.41166 7.768361 95.52021 105.3031 0.3412831 0.3303378 NA -0.4004154 22.27035 23.76433 NA 78.84012 24.96750 25.91641 NA 75.11838
7 MN99-71 FEG90-31 Yield 98.98546 5.298104 94.94591 103.0250 0.4959745 0.2776666 NA -0.1612311 21.86563 23.41697 NA 77.04669 24.85938 25.18159 NA 75.75153
11 MN96-141 FEG183-52 Yield 101.07137 5.571654 96.92884 105.2139 0.6132612 0.6961326 NA -0.4603399 24.36629 23.67733 NA 79.59345 27.88028 28.40806 NA 75.16009
15 MN99-02 FEG183-52 Yield 100.82967 9.499736 95.42053 106.2388 0.0249962 0.4211875 NA -0.1929275 26.15174 23.73807 NA 74.76672 26.29180 26.60344 NA 72.86270
19 FEG99-10 FEG148-56 Yield 98.01593 4.121802 94.45292 101.5789 0.5959335 0.6401323 NA -0.5080571 19.67552 19.58928 NA 85.81854 23.38585 24.23917 NA 80.69997
23 MN99-62 MN01-46 Yield 102.51094 4.673196 98.71709 106.3048 0.0755600 -0.0827658 NA 0.4010907 26.07342 28.37124 NA 72.67125 26.20580 28.16102 NA 74.51340

Benchmarking and comparisons

The code below compares the functions pop.predict and pop.predict2 with respect to computation time and results:

time1 <- system.time({
  capture.output(pop.predict.out <- pop.predict(
    G.in = G.in_ex_imputed, y.in = y.in_ex, map.in = map.in_ex, crossing.table = cross.tab_ex,
    nInd = 1000, nSim = 1, nCV.iter = 1, models = "rrBLUP"))
})

time2 <- system.time({pop.predict2.out <- pop.predict2(
  G.in = G.in_ex_imputed, y.in = y.in_ex, map.in = map.in_ex,
  crossing.table = cross.tab_ex,model = "rrBLUP")})

# Print the time (seconds) required for each function.
c(pop.predict = time1[[3]], pop.predict2 = time2[[3]])
#>  pop.predict pop.predict2 
#>        23.60         0.45

Plot results

predictions1 <- lapply(X = pop.predict.out$predictions, FUN = function(x) {
  x1 <- as.data.frame(apply(X = x, MARGIN = 2, FUN = unlist), stringsAsFactors = FALSE)
  cbind(x1[,c("Par1", "Par2")], sapply(X = x1[,-1:-2], as.numeric))
})

pop.predict.out1 <- predictions1$Yield_param.df[,c("Par1", "Par2", "pred.varG")]
pop.predict2.out1 <- subset(pop.predict2.out, trait == "Yield", c(parent1, parent2, pred_varG))

toplot <- merge(pop.predict.out1, pop.predict2.out1, by.x = c("Par1", "Par2"),
                by.y = c("parent1", "parent2"))

plot(pred.varG ~ pred_varG, toplot,
     xlab = "pop.predict2", ylab = "pop.predict",
     main = "Comparsion of predicted genetic variance")

Multi-parent populations

PopVar also includes functions for predicting the mean, genetic variance, and superior progeny mean of multi-parent populations. The mppop.predict function takes the same inputs as pop.predict or pop.predict2, and the mppop_predict2 function takes the same inputs as pop_predict2. Both require the additional argument n.parents, which will determine whether the populations are formed by 2- or 4-way matings (support for 8-way populations is forthcoming.)

Below is an example of using the mppop.predict function:

# Generate predictions for all possible 4-way crosses of 10 sample parents
sample_parents <- sample(unique(unlist(cross.tab_ex)), 10)

mp_out <- mppop.predict(G.in = G.in_ex_imputed, y.in = y.in_ex, map.in = map.in_ex,
                        parents = sample_parents, n.parents = 4, models = "rrBLUP")

knitr::kable(head(subset(mp_out, trait == "Yield")))
parent1 parent2 parent3 parent4 trait pred_mu pred_varG pred_musp_low pred_musp_high FHB DON Yield Height pred_cor_musp_low_FHB pred_cor_musp_low_DON pred_cor_musp_low_Yield pred_cor_musp_low_Height pred_cor_musp_high_FHB pred_cor_musp_high_DON pred_cor_musp_high_Yield pred_cor_musp_high_Height
3 FEG141-18 FEG148-56 FEG66-06 FEG69-38 Yield 98.40493 11.28119 92.51038 104.2995 0.4833239 0.5444313 NA -0.5193930 19.06880 19.25397 NA 82.43910 22.97332 23.63755 NA 77.23833
7 FEG141-18 FEG148-56 FEG66-06 FEG96-10 Yield 99.94246 10.31859 94.30501 105.5799 0.5540532 0.5974667 NA -0.4508954 17.88810 17.65336 NA 82.11881 22.24685 22.70391 NA 77.89269
11 FEG141-18 FEG148-56 FEG66-06 MN01-60 Yield 100.72094 11.78423 94.69641 106.7455 0.6228535 0.6181337 NA -0.5962415 18.82493 19.21492 NA 80.25695 24.36994 24.57592 NA 74.25710
15 FEG141-18 FEG148-56 FEG66-06 MN96-155 Yield 101.29767 13.99540 94.73221 107.8631 0.6418433 0.6859015 NA -0.5939293 19.28857 19.74097 NA 80.67510 25.14533 25.67244 NA 74.84535
19 FEG141-18 FEG148-56 FEG66-06 MN96-186 Yield 101.83161 11.53709 95.87058 107.7926 0.6502867 0.6718568 NA -0.6147056 18.62670 19.29492 NA 80.69938 24.34667 25.22889 NA 74.68221
23 FEG141-18 FEG148-56 FEG66-06 MN97-28 Yield 100.41829 9.53555 94.99895 105.8376 0.6111788 0.6572907 NA -0.5632576 18.92810 19.54156 NA 80.49933 24.15649 25.20063 NA 75.29882

Below is an example of using the mppop_predict2 function:

# Generate predictions for all possible 4-way crosses of 10 sample parents
mp_out2 <- mppop_predict2(M = G.in_ex_mat, y.in = y.in_ex, map.in = map.in_ex,
                          parents = sample_parents, n.parents = 4, models = "rrBLUP")

knitr::kable(head(subset(mp_out2, trait == "Yield")))
parent1 parent2 parent3 parent4 trait pred_mu pred_varG pred_musp_low pred_musp_high FHB DON Yield Height pred_cor_musp_low_FHB pred_cor_musp_low_DON pred_cor_musp_low_Yield pred_cor_musp_low_Height pred_cor_musp_high_FHB pred_cor_musp_high_DON pred_cor_musp_high_Yield pred_cor_musp_high_Height
3 FEG141-18 FEG148-56 FEG66-06 FEG69-38 Yield 98.40493 11.28119 92.51038 104.2995 0.4833239 0.5444313 NA -0.5193930 19.06880 19.25397 NA 82.43910 22.97332 23.63755 NA 77.23833
7 FEG141-18 FEG148-56 FEG66-06 FEG96-10 Yield 99.94246 10.31859 94.30501 105.5799 0.5540532 0.5974667 NA -0.4508954 17.88810 17.65336 NA 82.11881 22.24685 22.70391 NA 77.89269
11 FEG141-18 FEG148-56 FEG66-06 MN01-60 Yield 100.72094 11.78423 94.69641 106.7455 0.6228535 0.6181337 NA -0.5962415 18.82493 19.21492 NA 80.25695 24.36994 24.57592 NA 74.25710
15 FEG141-18 FEG148-56 FEG66-06 MN96-155 Yield 101.29767 13.99540 94.73221 107.8631 0.6418433 0.6859015 NA -0.5939293 19.28857 19.74097 NA 80.67510 25.14533 25.67244 NA 74.84535
19 FEG141-18 FEG148-56 FEG66-06 MN96-186 Yield 101.83161 11.53709 95.87058 107.7926 0.6502867 0.6718568 NA -0.6147056 18.62670 19.29492 NA 80.69938 24.34667 25.22889 NA 74.68221
23 FEG141-18 FEG148-56 FEG66-06 MN97-28 Yield 100.41829 9.53555 94.99895 105.8376 0.6111788 0.6572907 NA -0.5632576 18.92810 19.54156 NA 80.49933 24.15649 25.20063 NA 75.29882

References

Bernardo, Rex. 2010. Breeding for Quantitative Traits in Plants. 2nd ed. Woodbury, Minnesota: Stemma Press.

Mohammadi, Mohsen, Tyler Tiede, and Kevin P. Smith. 2015. “PopVar: A Genome-Wide Procedure for Predicting Genetic Variance and Correlated Response in Biparental Breeding Populations.” Crop Science 55 (5): 2068–77. https://doi.org/10.2135/cropsci2015.01.0030.

Schnell, F. W., and H. F. Utz. 1975. “F1-leistung und elternwahl euphyder züchtung von selbstbefruchtern.” In Bericht über Die Arbeitstagung Der Vereinigung Österreichischer Pflanzenzüchter, 243–48. Gumpenstein, Austria: BAL Gumpenstein.