The sommer package was developed to provide R users a powerful and reliable multivariate mixed model solver. The package is focused in problems of the type p > n (more effects to estimate than observations) and its core algorithm is coded in C++ using the Armadillo library. This package allows the user to fit mixed models with the advantage of specifying the variance-covariance structure for the random effects, and specify heterogeneous variances, and obtain other parameters such as BLUPs, BLUEs, residuals, fitted values, variances for fixed and random effects, etc.

The purpose of this quick start guide is to show the flexibility of the package under certain common scenarios:

B1) Background on mixed models

The core of the package is the mmer function which solve the mixed model equations. The functions are an interface to call the NR Direct-Inversion Newton-Raphson or Average Information algorithms (Tunnicliffe 1989; Gilmour et al. 1995; Lee et al. 2016). From version 2.0, sommer can handle multivariate models. Following Maier et al. (2015), the multivariate (and by extension the univariate) mixed model implemented has the form:

\(y_1 = X_1\beta_1 + Z_1u_1 + \epsilon_1\)

\(y_2 = X_2\beta_2 + Z_2u_2 + \epsilon_2\)

…

\(y_i = X_i\beta_i + Z_iu_i + \epsilon_i\)

where \(y_i\) is a vector of trait phenotypes, \(\beta_i\) is a vector of fixed effects, \(u_i\) is a vector of random effects for individuals and \(e_i\) are residuals for trait ‘i’ (i = 1, …, t). The random effects (\(u_1\) … \(u_i\) and \(e_i\)) are assumed to be normally distributed with mean zero. X and Z are incidence matrices for fixed and random effects respectively. The distribution of the multivariate response and the phenotypic variance covariance (V) are:

\(Y = X\beta + ZU + \epsilon_i\)

Y ~ MVN(\(X\beta\), V)

\[\mathbf{Y} = \left[\begin{array} {r} y_1 \\ y_2 \\ ... \\ y_t \\ \end{array}\right] \]

\[\mathbf{X} = \left[\begin{array} {rrr} X_1 & 0 & 0 \\ \vdots & \ddots & \vdots\\ 0 & 0 & X_t \\ \end{array}\right] \]

\[\mathbf{V} = \left[\begin{array} {rrr} Z_1 K{\sigma^2_{g_{1}}} Z_1' + H{\sigma^2_{\epsilon_{1}}} & ... & Z_1 K{\sigma_{g_{1,t}}} Z_t' + H{\sigma_{\epsilon_{1,t}}} \\ \vdots & \ddots & \vdots\\ Z_1 K{\sigma_{g_{1,t}}} Z_t' + H{\sigma_{\epsilon_{1,t}}} & ... & Z_t K{\sigma^2_{g_{t}}} Z_t' + H{\sigma^2_{\epsilon_{t}}} \\ \end{array}\right] \]

where K is the relationship or covariance matrix for the kth random effect (u=1,…,k), and H=I is an identity matrix or a partial identity matrix for the residual term. The terms \(\sigma^2_{g_{i}}\) and \(\sigma^2_{\epsilon_{i}}\) denote the genetic (or any of the kth random terms) and residual variance of trait ‘i’, respectively and \(\sigma_{g_{_{ij}}}\) and \(\sigma_{\epsilon_{_{ij}}}\) the genetic (or any of the kth random terms) and residual covariance between traits ‘i’ and ‘j’ (i=1,…,t, and j=1,…,t). The algorithm implemented optimizes the log likelihood:

\(logL = 1/2 * ln(|V|) + ln(X'|V|X) + Y'PY\)

where || is the determinant of a matrix. And the REML estimates are updated using a Newton optimization algorithm of the form:

\(\theta^{k+1} = \theta^{k} + (H^{k})^{-1}*\frac{dL}{d\sigma^2_i}|\theta^k\)

Where, \(\theta\) is the vector of variance components for random effects and covariance components among traits, \(H^{-1}\) is the inverse of the Hessian matrix of second derivatives for the kth cycle, \(\frac{dL}{d\sigma^2_i}\) is the vector of first derivatives of the likelihood with respect to the variance-covariance components. The Eigen decomposition of the relationship matrix proposed by Lee and Van Der Werf (2016) was included in the Newton-Raphson algorithm to improve time efficiency. Additionally, the popular pin function to estimate standard errors for linear combinations of variance components (i.e. heritabilities and genetic correlations) was added to the package as well.

Please refer to the canonical papers listed in the Literature section to check how the algorithms work. We have tested widely the methods to make sure they provide the same solution when the likelihood behaves well but for complex problems they might lead to slightly different answers. If you have any concern please contact me at cova_ruber@live.com.mx.

In the following section we will go in detail over several examples on how to use mixed models in univariate and multivariate case and their use in quantitative genetics.

B2) Background on covariance structures

One of the major strenghts of linear mixed models is the flexibility to specify variance-covariance structures at all levels. In general, variance structures of mixed models can be seen as tensor (kronecker) products of multiple variance-covariance stuctures. For example, a multi-response model (i.e. 2 traits) where “g” individuals (i.e. 100 individuals) are tested in “e” treatments (i.e. 3 environments), the variance-covariance for the random effect “individuals” can be seen as the following multiplicative model:

T \(\otimes\) G \(\otimes\) A

where:

\[\mathbf{T} = \left[\begin{array} {rr} {\sigma^2_{g_{_{t1,t1}}}} & {\sigma_{g_{_{t1,t2}}}} \\ {\sigma_{g_{_{t2,t1}}}} & {\sigma^2_{g_{_{t2,t2}}}} \\ \end{array}\right] \]

is the covariance structure for individuals among traits.

\[\mathbf{G} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & {\sigma_{g_{_{e1,e2}}}} & {\sigma_{g_{_{e1,e3}}}} \\ {\sigma_{g_{_{e2,e1}}}} & {\sigma^2_{g_{_{e2,e2}}}} & {\sigma_{g_{_{e2,e3}}}} \\ {\sigma_{g_{_{e3,e1}}}} & {\sigma_{g_{_{e3,e2}}}} & {\sigma^2_{g_{_{e3,e3}}}} \\ \end{array}\right] \]

is the covariance structure for individuals among environments.

and \(A\) is a square matrix representing the covariance among the levels of the individuals (any known relationship matrix).

The T and G covariance structures shown above are unknown matrices to be estimated whereas A is known. The T and G matrices shown above are called as unstructured (US) covariance matrices, although this type is just one example from several covariance structures that the linear mixed models enable. For example, other popular covariance structures are:

Diagonal (DIAG) covariance structures

\[\mathbf{\Sigma} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & 0 & 0 \\ \vdots & \ddots & \vdots \\ 0 & 0 & {\sigma^2_{g_{_{ei,ei}}}} \\ \end{array}\right] \] Compound simmetry (CS) covariance structures

\[\mathbf{\Sigma} = \left[\begin{array} {rrrr} {\sigma^2_{g}} + {\sigma^2_{ge}} & {\sigma^2_{g}} & {\sigma^2_{g}} & {\sigma^2_{g}} \\ {\sigma^2_{g}} & {\sigma^2_{g}} + {\sigma^2_{ge}} & {\sigma^2_{g}} & {\sigma^2_{g}}\\ \vdots & \vdots & \ddots & \vdots\\ {\sigma^2_{g}} & {\sigma^2_{g}} & {\sigma^2_{g}} & {\sigma^2_{g}} + {\sigma^2_{ge}}\\ \end{array}\right] \]

First order autoregressive (AR1) covariance structures

\[\mathbf{\Sigma} = \sigma^2 \left[\begin{array} {rrrr} 1 & {\rho} & {\rho^2} & {\rho^3} \\ {\rho} & 1 & {\rho} & {\rho^2}\\ {\rho^2} & {\rho} & 1 & {\rho} \\ {\rho^3} & {\rho^2} & {\rho} & 1 \\ \end{array}\right] \]

or the already mentioned Unstructured (US) covariance structures

\[\mathbf{\Sigma} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & {\sigma_{g_{_{e1,e2}}}} & {\sigma_{g_{_{e1,e3}}}} \\ \vdots & \ddots & \vdots \\ {\sigma_{g_{_{e3,e1}}}} & {\sigma_{g_{_{e3,e2}}}} & {\sigma^2_{g_{_{e3,e3}}}} \\ \end{array}\right] \]

among others. Sommer has the capabilities to fit some of these covariance structures in the mixed model machinery.

1) Univariate homogeneous variance models

This type of models refer to single response models where a variable of interest (i.e. genotypes) needs to be analized as interacting with a 2nd random effect (i.e. environments), but you assume that across environments the genotypes have the same variance component. This is the so-called compound simmetry (CS) model.

library(sommer)
data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

ans1 <- mmer(Yield~Env,
              random= ~ Name + Env:Name,
              rcov= ~ units,
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -31.2668   11:5:53      0           0
##     2      -23.2804   11:5:53      0           0
##     3      -20.4746   11:5:53      0           0
##     4      -20.1501   11:5:53      0           0
##     5      -20.1454   11:5:53      0           0
##     6      -20.1454   11:5:53      0           0

summary(ans1)

## ============================================================
##          Multivariate Linear Mixed Model fit by REML         
## **********************  sommer 4.0  ********************** 
## ============================================================
##          logLik      AIC      BIC Method Converge
## Value -20.14538 46.29075 55.95182     NR     TRUE
## ============================================================
## Variance-Covariance components:
##                      VarComp VarCompSE Zratio Constraint
## Name.Yield-Yield       3.682     1.691  2.177   Positive
## Env:Name.Yield-Yield   5.173     1.495  3.460   Positive
## units.Yield-Yield      4.366     0.647  6.748   Positive
## ============================================================
## Fixed effects:
##   Trait      Effect Estimate Std.Error t.value
## 1 Yield (Intercept)   16.496    0.6855  24.065
## 2 Yield  EnvCA.2012   -5.777    0.7558  -7.643
## 3 Yield  EnvCA.2013   -6.380    0.7960  -8.015
## ============================================================
## Groups and observations:
##          Yield
## Name        41
## Env:Name   123
## ============================================================
## Use the '$' sign to access results and parameters

2) Univariate heterogeneous variance models

Very often in multi-environment trials, the assumption that the genetic variance or the residual variance is the same across locations may be too naive. Because of that, specifying a general genetic component and a location specific genetic variance is the way to go. This requires a CS+DIAG model (also called heterogeneous CS model).

data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

ans2 <- mmer(Yield~Env,
              random= ~Name + vs(ds(Env),Name),
              rcov= ~ vs(ds(Env),units),
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -31.2668   11:5:53      0           0
##     2      -19.8549   11:5:53      0           0
##     3      -15.9797   11:5:53      0           0
##     4      -15.4374   11:5:53      0           0
##     5      -15.43   11:5:53      0           0
##     6      -15.4298   11:5:53      0           0

summary(ans2)

## ============================================================
##          Multivariate Linear Mixed Model fit by REML         
## **********************  sommer 4.0  ********************** 
## ============================================================
##          logLik      AIC      BIC Method Converge
## Value -15.42983 36.85965 46.52072     NR     TRUE
## ============================================================
## Variance-Covariance components:
##                           VarComp VarCompSE Zratio Constraint
## Name.Yield-Yield            2.963     1.496  1.980   Positive
## CA.2011:Name.Yield-Yield   10.146     4.507  2.251   Positive
## CA.2012:Name.Yield-Yield    1.878     1.870  1.004   Positive
## CA.2013:Name.Yield-Yield    6.629     2.503  2.649   Positive
## CA.2011:units.Yield-Yield   4.942     1.525  3.242   Positive
## CA.2012:units.Yield-Yield   5.725     1.312  4.363   Positive
## CA.2013:units.Yield-Yield   2.560     0.640  4.000   Positive
## ============================================================
## Fixed effects:
##   Trait      Effect Estimate Std.Error t.value
## 1 Yield (Intercept)   16.508    0.8268  19.965
## 2 Yield  EnvCA.2012   -5.817    0.8575  -6.783
## 3 Yield  EnvCA.2013   -6.412    0.9356  -6.854
## ============================================================
## Groups and observations:
##              Yield
## Name            41
## CA.2011:Name    41
## CA.2012:Name    41
## CA.2013:Name    41
## ============================================================
## Use the '$' sign to access results and parameters

As you can see the special function at or diag can be used to indicate that there’s a different variance for the genotypes in each environment. Same was done for the residual. The difference between at and diag is that the at function can be used to specify the levels or specific environments where the variance is different.

3) Unstructured variance models

A more relaxed asumption than the CS+DIAG model is the unstructured model (US) which assumes that among the levels of certain factor (i.e. Environments) there’s a covariance struture of a second random effect (i.e. Genotypes). This can be done in sommer using the us(.) function:

data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

ans3 <- mmer(Yield~Env,
             random=~ vs(us(Env),Name),
             rcov=~vs(us(Env),units), 
             data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -37.9059   11:5:53      0           0
##     2      -17.9745   11:5:53      0           0
##     3      -12.2427   11:5:54      1           0
##     4      -11.5121   11:5:54      1           0
##     5      -11.5001   11:5:54      1           0
##     6      -11.4997   11:5:54      1           0

summary(ans3)

## ===================================================================
##              Multivariate Linear Mixed Model fit by REML             
## **************************  sommer 4.0  ************************** 
## ===================================================================
##          logLik      AIC      BIC Method Converge
## Value -11.49971 28.99943 38.66049     NR     TRUE
## ===================================================================
## Variance-Covariance components:
##                                   VarComp VarCompSE    Zratio Constraint
## CA.2011:Name.Yield-Yield           15.665 5.421e+00 2.890e+00   Positive
## CA.2012:CA.2011:Name.Yield-Yield    6.110 2.485e+00 2.459e+00   Unconstr
## CA.2012:Name.Yield-Yield            4.530 1.821e+00 2.488e+00   Positive
## CA.2013:CA.2011:Name.Yield-Yield    6.384 3.066e+00 2.082e+00   Unconstr
## CA.2013:CA.2012:Name.Yield-Yield    0.393 1.523e+00 2.580e-01   Unconstr
## CA.2013:Name.Yield-Yield            8.597 2.484e+00 3.461e+00   Positive
## CA.2011:units.Yield-Yield           4.970 1.532e+00 3.243e+00   Positive
## CA.2012:CA.2011:units.Yield-Yield   4.087 2.436e-16 1.678e+16   Unconstr
## CA.2012:units.Yield-Yield           5.673 1.301e+00 4.361e+00   Positive
## CA.2013:CA.2011:units.Yield-Yield   4.087 0.000e+00       Inf   Unconstr
## CA.2013:CA.2012:units.Yield-Yield   4.087 0.000e+00       Inf   Unconstr
## CA.2013:units.Yield-Yield           2.557 6.393e-01 4.000e+00   Positive
## ===================================================================
## Fixed effects:
##   Trait      Effect Estimate Std.Error t.value
## 1 Yield (Intercept)   16.331    0.8137  20.070
## 2 Yield  EnvCA.2012   -5.696    0.7404  -7.693
## 3 Yield  EnvCA.2013   -6.271    0.8191  -7.656
## ===================================================================
## Groups and observations:
##                      Yield
## CA.2011:Name            41
## CA.2012:CA.2011:Name    82
## CA.2012:Name            41
## CA.2013:CA.2011:Name    82
## CA.2013:CA.2012:Name    82
## CA.2013:Name            41
## ===================================================================
## Use the '$' sign to access results and parameters

As can be seen the us(Env) indicates that the genotypes (Name) can have a covariance structure among environments (Env).

4) Multivariate homogeneous variance models

Currently there’s a great push for multi-response models. This is motivated by the correlation that certain variables hide and that could benefit in the prediction perspective. In sommer to specify multivariate models the response requires the use of the cbind() function in the response, and the us(trait), diag(trait), or at(trait) functions in the random part of the model.

data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

DT$EnvName <- paste(DT$Env,DT$Name)
ans4 <- mmer(cbind(Yield, Weight) ~ Env,
              random= ~ vs(Name, Gtc=unsm(2)) + vs(EnvName, Gtc=unsm(2)),
              rcov= ~ vs(units, Gtc=unsm(2)),
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      66.0395   11:5:54      0           0
##     2      131.529   11:5:55      1           0
##     3      162.769   11:5:55      1           0
##     4      166.983   11:5:56      2           0
##     5      167.025   11:5:57      3           0
##     6      167.025   11:5:57      3           0

summary(ans4)

## ============================================================
##          Multivariate Linear Mixed Model fit by REML         
## **********************  sommer 4.0  ********************** 
## ============================================================
##         logLik       AIC       BIC Method Converge
## Value 167.0252 -322.0505 -298.5695     NR     TRUE
## ============================================================
## Variance-Covariance components:
##                         VarComp VarCompSE Zratio Constraint
## u:Name.Yield-Yield       3.7089   1.68117  2.206   Positive
## u:Name.Yield-Weight      0.9071   0.37944  2.391   Unconstr
## u:Name.Weight-Weight     0.2243   0.08775  2.557   Positive
## u:EnvName.Yield-Yield    5.0921   1.47879  3.443   Positive
## u:EnvName.Yield-Weight   1.0269   0.30767  3.338   Unconstr
## u:EnvName.Weight-Weight  0.2101   0.06661  3.154   Positive
## u:units.Yield-Yield      4.3837   0.64941  6.750   Positive
## u:units.Yield-Weight     0.9077   0.14145  6.417   Unconstr
## u:units.Weight-Weight    0.2280   0.03377  6.751   Positive
## ============================================================
## Fixed effects:
##    Trait      Effect Estimate Std.Error t.value
## 1  Yield (Intercept)  16.4093    0.6783  24.191
## 2 Weight (Intercept)   0.9806    0.1497   6.550
## 3  Yield  EnvCA.2012  -5.6844    0.7474  -7.606
## 4 Weight  EnvCA.2012  -1.1846    0.1593  -7.439
## 5  Yield  EnvCA.2013  -6.2952    0.7850  -8.019
## 6 Weight  EnvCA.2013  -1.3559    0.1681  -8.065
## ============================================================
## Groups and observations:
##           Yield Weight
## u:Name       41     41
## u:EnvName    94     94
## ============================================================
## Use the '$' sign to access results and parameters

You may notice that we have added the us(trait) behind the random effects. This is to indicate the structure that should be assume in the multivariate model. The diag(trait) used behind a random effect (i.e. Name) indicates that for the traits modeled (Yield and Weight) there’s no a covariance component and should not be estimated, whereas us(trait) assumes that for such random effect, there’s a covariance component to be estimated (i.e. covariance between Yield and Weight for the random effect Name). Same applies for the residual part (rcov).

5) Multivariate heterogeneous variance models

This is just an extension of the univariate heterogeneous variance models but at the multivariate level. This would be a CS+DIAG multivariate model:

data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

DT$EnvName <- paste(DT$Env,DT$Name)
ans5 <- mmer(cbind(Yield, Weight) ~ Env,
              random= ~ vs(Name, Gtc=unsm(2)) + vs(ds(Env),Name, Gtc=unsm(2)),
              rcov= ~ vs(ds(Env),units, Gtc=unsm(2)),
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      66.0395   11:5:58      1           0
##     2      138.617   11:5:59      2           0
##     3      172.682   11:6:0      3           0
##     4      177.662   11:6:1      4           0
##     5      177.801   11:6:2      5           0
##     6      177.813   11:6:4      7           0
##     7      177.815   11:6:5      8           0
##     8      177.815   11:6:6      9           0

summary(ans5)

## =============================================================
##           Multivariate Linear Mixed Model fit by REML          
## **********************  sommer 4.0  ********************** 
## =============================================================
##         logLik       AIC       BIC Method Converge
## Value 177.8154 -343.6308 -320.1497     NR     TRUE
## =============================================================
## Variance-Covariance components:
##                             VarComp VarCompSE Zratio Constraint
## u:Name.Yield-Yield          3.31936   1.45269 2.2850   Positive
## u:Name.Yield-Weight         0.79393   0.32621 2.4338   Unconstr
## u:Name.Weight-Weight        0.19085   0.07503 2.5438   Positive
## CA.2011:Name.Yield-Yield    8.70657   4.01470 2.1687   Positive
## CA.2011:Name.Yield-Weight   1.77892   0.83926 2.1196   Unconstr
## CA.2011:Name.Weight-Weight  0.35966   0.17903 2.0089   Positive
## CA.2012:Name.Yield-Yield    2.57109   1.94951 1.3188   Positive
## CA.2012:Name.Yield-Weight   0.33245   0.39840 0.8345   Unconstr
## CA.2012:Name.Weight-Weight  0.03842   0.08595 0.4470   Positive
## CA.2013:Name.Yield-Yield    5.46908   2.16307 2.5284   Positive
## CA.2013:Name.Yield-Weight   1.34713   0.50479 2.6687   Unconstr
## CA.2013:Name.Weight-Weight  0.32902   0.12208 2.6952   Positive
## CA.2011:units.Yield-Yield   4.93852   1.52318 3.2422   Positive
## CA.2011:units.Yield-Weight  0.99447   0.32150 3.0932   Unconstr
## CA.2011:units.Weight-Weight 0.23982   0.07394 3.2433   Positive
## CA.2012:units.Yield-Yield   5.73887   1.31533 4.3631   Positive
## CA.2012:units.Yield-Weight  1.28009   0.30157 4.2448   Unconstr
## CA.2012:units.Weight-Weight 0.31806   0.07286 4.3652   Positive
## CA.2013:units.Yield-Yield   2.56127   0.63993 4.0024   Positive
## CA.2013:units.Yield-Weight  0.44569   0.12645 3.5246   Unconstr
## CA.2013:units.Weight-Weight 0.12232   0.03057 4.0009   Positive
## =============================================================
## Fixed effects:
##    Trait      Effect Estimate Std.Error t.value
## 1  Yield (Intercept)  16.4243    0.7891  20.815
## 2 Weight (Intercept)   0.9866    0.1683   5.863
## 3  Yield  EnvCA.2012  -5.7339    0.8266  -6.937
## 4 Weight  EnvCA.2012  -1.1998    0.1698  -7.066
## 5  Yield  EnvCA.2013  -6.3128    0.8757  -7.209
## 6 Weight  EnvCA.2013  -1.3621    0.1915  -7.114
## =============================================================
## Groups and observations:
##              Yield Weight
## u:Name          41     41
## CA.2011:Name    41     41
## CA.2012:Name    41     41
## CA.2013:Name    41     41
## =============================================================
## Use the '$' sign to access results and parameters

6) Multivariate unstructured variance models

This is just an extension of the univariate unstructured variance models but at the multivariate level. This would be a US multivariate model:

data(DT_example)
head(DT)

##                   Name     Env Loc Year     Block Yield    Weight
## 33  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.1     4 -1.904711
## 65          CO02024-9W CA.2013  CA 2013 CA.2013.1     5 -1.446958
## 66  Manistee(MSL292-A) CA.2013  CA 2013 CA.2013.2     5 -1.516271
## 67            MSL007-B CA.2011  CA 2011 CA.2011.2     5 -1.435510
## 68           MSR169-8Y CA.2013  CA 2013 CA.2013.1     5 -1.469051
## 103         AC05153-1W CA.2013  CA 2013 CA.2013.1     6 -1.307167

DT$EnvName <- paste(DT$Env,DT$Name)
ans6 <- mmer(cbind(Yield, Weight) ~ Env,
              random= ~ vs(us(Env),Name, Gtc=unsm(2)),
              rcov= ~ vs(ds(Env),units, Gtc=unsm(2)),
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      56.6189   11:6:7      1           0
##     2      140.894   11:6:8      2           0
##     3      176.238   11:6:10      4           0
##     4      181.462   11:6:11      5           0
##     5      181.688   11:6:12      6           0
##     6      181.746   11:6:14      8           0
##     7      181.77   11:6:15      9           0
##     8      181.781   11:6:17      11           0
##     9      181.787   11:6:18      12           0
##     10      181.791   11:6:19      13           0
##     11      181.793   11:6:21      15           0
##     12      181.794   11:6:22      16           0
##     13      181.794   11:6:23      17           0

summary(ans6)

## ====================================================================
##              Multivariate Linear Mixed Model fit by REML             
## **************************  sommer 4.0  ************************** 
## ====================================================================
##         logLik       AIC       BIC Method Converge
## Value 181.7945 -351.5889 -328.1079     NR     TRUE
## ====================================================================
## Variance-Covariance components:
##                                    VarComp VarCompSE Zratio Constraint
## CA.2011:Name.Yield-Yield           15.6451   5.35692  2.921   Positive
## CA.2011:Name.Yield-Weight           3.3586   1.14633  2.930   Unconstr
## CA.2011:Name.Weight-Weight          0.7182   0.24871  2.888   Positive
## CA.2012:CA.2011:Name.Yield-Yield    6.5289   2.48615  2.626   Positive
## CA.2012:CA.2011:Name.Yield-Weight   1.3505   0.52388  2.578   Unconstr
## CA.2012:CA.2011:Name.Weight-Weight  0.2842   0.11259  2.524   Positive
## CA.2012:Name.Yield-Yield            4.7893   1.86183  2.572   Positive
## CA.2012:Name.Yield-Weight           0.8640   0.38377  2.251   Unconstr
## CA.2012:Name.Weight-Weight          0.1693   0.08354  2.027   Positive
## CA.2013:CA.2011:Name.Yield-Yield    5.9934   2.93830  2.040   Positive
## CA.2013:CA.2011:Name.Yield-Weight   1.4232   0.64973  2.190   Unconstr
## CA.2013:CA.2011:Name.Weight-Weight  0.3379   0.14680  2.302   Positive
## CA.2013:CA.2012:Name.Yield-Yield    2.0987   1.44034  1.457   Positive
## CA.2013:CA.2012:Name.Yield-Weight   0.5240   0.32356  1.619   Unconstr
## CA.2013:CA.2012:Name.Weight-Weight  0.1342   0.07572  1.772   Positive
## CA.2013:Name.Yield-Yield            8.6257   2.47811  3.481   Positive
## CA.2013:Name.Yield-Weight           2.1048   0.58748  3.583   Unconstr
## CA.2013:Name.Weight-Weight          0.5125   0.14285  3.588   Positive
## CA.2011:units.Yield-Yield           4.9516   1.52694  3.243   Positive
## CA.2011:units.Yield-Weight          0.9993   0.32286  3.095   Unconstr
## CA.2011:units.Weight-Weight         0.2411   0.07432  3.244   Positive
## CA.2012:units.Yield-Yield           5.7790   1.32423  4.364   Positive
## CA.2012:units.Yield-Weight          1.2914   0.30408  4.247   Unconstr
## CA.2012:units.Weight-Weight         0.3212   0.07356  4.366   Positive
## CA.2013:units.Yield-Yield           2.5567   0.63883  4.002   Positive
## CA.2013:units.Yield-Weight          0.4452   0.12631  3.524   Unconstr
## CA.2013:units.Weight-Weight         0.1223   0.03056  4.001   Positive
## ====================================================================
## Fixed effects:
##    Trait      Effect Estimate Std.Error t.value
## 1  Yield (Intercept)  16.3342    0.8254  19.790
## 2 Weight (Intercept)   0.9677    0.1770   5.466
## 3  Yield  EnvCA.2012  -5.6637    0.7449  -7.604
## 4 Weight  EnvCA.2012  -1.1855    0.1604  -7.390
## 5  Yield  EnvCA.2013  -6.2153    0.8340  -7.453
## 6 Weight  EnvCA.2013  -1.3406    0.1806  -7.425
## ====================================================================
## Groups and observations:
##                      Yield Weight
## CA.2011:Name            41     41
## CA.2012:CA.2011:Name    82     82
## CA.2012:Name            41     41
## CA.2013:CA.2011:Name    82     82
## CA.2013:CA.2012:Name    82     82
## CA.2013:Name            41     41
## ====================================================================
## Use the '$' sign to access results and parameters

Any number of random effects can be specified with different structures.

7) Details on special functions for variance models

the major vs() function for special variance models and its auxiliars

The sommer function vs() allows to construct complex variance models that are passed to the mmer() function it constitutes one of the most important features of the sommer package. Its specification of the vs() function has the form:

random=~vs(..., Gu, Gt, Gtc)

The idea is that the vs() function reflects the special variance structure that each random effect could have in the matrix notation:

\(var(u) = T \bigotimes E \bigotimes ... \bigotimes A\)

where the … argument in the vs() function is used to specify the kronecker products from all matrices that form the variance for the random effect , where the auxiliar function ds(), us(), cs(), at(), can be used to define such structure variance structure. The idea is that a variance model for a random effect x (i.e. individuals) might require a more flexible model than just:

random=~x

For example, if individuals are tested in different time-points and environment, we can assume a different variance and covariance components among the individuals in the different environment-timepoint combinations. An example of variance structure of the type:

\(var(u) = T \bigotimes E \bigotimes S \bigotimes A\)

would be specified in the vs() function as:

random=~vs(us(e),us(s),x, Gu=A, Gtc=T)

where the e would be a column vector in a data frame for the environments, s a column vector in the dataframe for the time points, x is the vector in the datrame for the identifier of individuals, A is a known square variance covariance matrix among individuals (usually an identity matrix; default if not specified), and T is a square matrices with as many rows and columns as the number of traits that specifyies the trait covariance structure.

The auxiliar function to build the variance models for the random effect are: + ds() diagonal covariance structure + us() unstructured covariance + at() specific levels heterogeneous variance structure + cs() customized covariance structure

ds() to specify a diagonal (DIAG) covariance structures

A diagonal covariance structure looks like this:

\[\mathbf{\Sigma} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & 0 & 0 \\ \vdots & \ddots & \vdots \\ 0 & 0 & {\sigma^2_{g_{_{ei,ei}}}} \\ \end{array}\right] \] Considering an example for one random effect (g; indicating i.e. individuals) evaluated in different treatment levels (e; indicating i.e. the different treatments) the model would look like:

random=~vs(ds(e),g)

us() to specify an unstructured (US) covariance

A unstructured covariance looks like this:

Considering same example for one random effect (g; indicating i.e. individuals) evaluated in different treatment levels (e; indicating i.e. the different treatments) the model would look like:

random=~vs(us(e),g)

at() to specify a level-specific heterogeneous variance

A diagonal covariance structure for specific levels of the second random effect looks like this:

\[\mathbf{\Sigma} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & 0 & 0 \\ \vdots & \ddots & \vdots \\ 0 & 0 & {\sigma^2_{g_{_{ei,ei}}}} \\ \end{array}\right] \]

Considering same example for one random effect (g; indicating i.e. individuals) evaluated in different treatment levels (e; indicating i.e. the different treatments A,B,C) the model would look like:

random=~vs(at(e,c("A","B")),g)

where the variance component for g is only fitted at levels A and B.

cs() to specify a level-specific variance-covariance structure

A customized covariance structure for specific levels of the second random effect (variance and covariances) looks i.e. like this:

\[\mathbf{\Sigma} = \left[\begin{array} {rrr} {\sigma^2_{g_{_{e1,e1}}}} & {\sigma_{g_{_{e1,e2}}}} & 0 \\ \vdots & \ddots & \vdots \\ 0 & 0 & {\sigma^2_{g_{_{ei,ei}}}} \\ \end{array}\right] \]

Considering same example for one random effect (g; indicating i.e. individuals) evaluated in different treatment levels (e; indicating i.e. the different treatments A,B,C) the model would look like:

random=~vs(cs(e,mm),g)

where mm indicates which variance and covariance components are estimated for g.

8) The specification of constraints in the variance components (Gtc argument)

One of the major strengths of sommer is its extreme flexibility to specify variance-covariance structures in the multi-trait framework. Since sommer 3.7 this is easily achieved by the use of the vs() function and it’s argument Gtc. The Gtc argument expects a matrix of constraints for the variance-covariance components for the random effect filled with numbers according to the following rules:

0: parameter not to be estimated 1: estimated and constrained to be positive 2: estimated and unconstrained 3: not to be estimated but fixed value provided in Gt

Some useful function to specify quickly the contraint matrices are unsm() for unstructured, uncm for unconstrained, fixm() for fixed constraint, and fcm() for fixed effect constrains.

Consider a multi-trait model with 4 traits (y1,…,y4) and 1 random effects (u) and 1 fixed effect (x)

fixed=cbind(y1,y2,y3,y4)~x

random= ~vs(u, Gtc=?)

The constraint for the 4 x 4 matrix of variance covariance components to be estimated can be an:

unstructured (variance components have to be positive and covariances either positive or negative) random= ~vs(u, Gtc=unsm(4))

unsm(4)

##      [,1] [,2] [,3] [,4]
## [1,]    1    2    2    2
## [2,]    2    1    2    2
## [3,]    2    2    1    2
## [4,]    2    2    2    1

unconstrained (any component variance or covariance can be positive or negative) random= ~vs(u, Gtc=uncm(4))

uncm(4)

##      [,1] [,2] [,3] [,4]
## [1,]    2    2    2    2
## [2,]    2    2    2    2
## [3,]    2    2    2    2
## [4,]    2    2    2    2

fixed (variance or covariance components indicated with a 3 are considered fixed and values are provided in the Gt argument) random= ~vs(u, Gtc=fixm(4), Gt=mm)

fixm(4)

##      [,1] [,2] [,3] [,4]
## [1,]    3    3    3    3
## [2,]    3    3    3    3
## [3,]    3    3    3    3
## [4,]    3    3    3    3

where mm is a 4 x 4 matrix with initial values for the variance components.

constraints for fixed effects fixed= cbind(y1,y2,y3,y4)~vs(x, Gtc=fcm(c(1,0,1,0)))

fcm(c(1,0,1,0))

##      [,1] [,2]
## [1,]    1    0
## [2,]    0    0
## [3,]    0    1
## [4,]    0    0

where 1’s and 0’s indicate the traits where the fixed effect will be estimated (1’s) and where it won’t (0’s).

9) Special functions for special models

Random regression models

In order to fit random regression models the user can use the leg() function to fit Legendre polynomials. This can be combined with other special covariance structures such as ds(), us(), etc.

library(orthopolynom)

## Loading required package: polynom

data(DT_legendre)
head(DT)

##     SUBJECT X          Y Xf
## 1.1       1 1 -0.7432795  1
## 2.1       2 1 -0.6669945  1
## 3.1       3 1 -4.2802751  1
## 4.1       4 1  4.1092149  1
## 5.1       5 1 -3.0317213  1
## 6.1       6 1  1.3506577  1

mRR2<-mmer(Y~ 1 + Xf
           , random=~ vs(us(leg(X,1)),SUBJECT)
           , rcov=~vs(units)
           , data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -145.279   11:6:24      0           0
##     2      -138.353   11:6:24      0           0
##     3      -136.403   11:6:24      0           0
##     4      -136.224   11:6:24      0           0
##     5      -136.222   11:6:25      1           0
##     6      -136.222   11:6:25      1           0

summary(mRR2)$varcomp

##                         VarComp VarCompSE   Zratio Constraint
## leg0:SUBJECT.Y-Y      2.5782969 0.6717242 3.838326   Positive
## leg1:leg0:SUBJECT.Y-Y 0.4765431 0.2394975 1.989763   Unconstr
## leg1:SUBJECT.Y-Y      0.3497299 0.2183229 1.601893   Positive
## u:units.Y-Y           2.6912226 0.3825197 7.035513   Positive

Here, a numeric covariate X is used to explain the trajectory of the SUBJECT’s and combined with an unstructured covariance matrix. The details can be found in the theory.

GWAS models

Although genome wide association studies can be conducted through a variety of approaches, the use of mixed models to find association between markers and phenotypes still one of the most popular approaches. Two of the most classical and popular approaches is to test marker by marker trough mixed modeling (1 model by marker) to obtain the marker effect and an statistic reflecting the level of association usually provided as the -log10 p-value. The second most popular approach is to assume that the genetic variance component is similar for all markers and therefore the variance components are only estimated once (1 model for all markers) and use the inverse of the phenotypic variance matrix (V.inverse) to test all markers in the generalized linear model b=(XV-X)-XV-y. This makes the GWAS much faster and efficient without major loses. Given the straight forward extension, sommer provides the GWAS function which can fit both type of approaches (be aware that these are 2 among many existant in the literature) in univariate and multivariate models, that way genetically correlated traits can be tested together to increase the power of detection. In summary the GWAS model implemented in sommer to obtain marker effect is a generalized linear model of the form:

b = (X’V-X)X’V-y

with X = ZMi

where: b is the marker effect (dimensions 1 x mt) y is the response variable (univariate or multivariate) (dimensions 1 x nt) V- is the inverse of the phenotypic variance matrix (dimensions nt x nt) Z is the incidence matrix for the random effect selected (gTerm argument) to perform the GWAS (dimensions nt x ut) Mi is the ith column of the marker matrix (M argument) (dimensions u x m)

for t traits, n observations, m markers and u levels of the random effect. Depending if P3D is TRUE or FALSE the V- matrix will be calculated once and used for all marker tests (P3D=TRUE) or estimated through REML for each marker (P3D=FALSE).

Here we show a simple GWAS model for an univariate example.

data(DT_cpdata)
#### create the variance-covariance matrix
A <- A.mat(GT) # additive relationship matrix
#### look at the data and fit the model
head(DT,3)

##        id Row Col Year      color  Yield FruitAver Firmness Rowf Colf
## P003 P003   3   1 2014 0.10075269 154.67     41.93  588.917    3    1
## P004 P004   4   1 2014 0.13891940 186.77     58.79  640.031    4    1
## P005 P005   5   1 2014 0.08681502  80.21     48.16  671.523    5    1

head(MP,3)

##                 Locus Position Chrom
## 1  scaffold_77830_839        0     1
## 2  scaffold_39187_895        0     1
## 3 scaffold_50439_2379        0     1

GT[1:3,1:4]

##      scaffold_50439_2381 scaffold_39344_153 uneak_3436043 uneak_2632033
## P003                   0                  0             0             1
## P004                   0                  0             0             1
## P005                   0                 -1             0             1

mix1 <- GWAS(color~1,
             random=~vs(id,Gu=A)
             + Rowf + Colf,
             rcov=~units,
             data=DT,
             M=GT, gTerm = "u:id")

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -143.207   11:6:26      0           0
##     2      -117.977   11:6:26      0           0
##     3      -109.877   11:6:27      1           1
##     4      -108.178   11:6:27      1           1
##     5      -108.123   11:6:27      1           1
##     6      -108.12   11:6:28      2           1
##     7      -108.12   11:6:28      2           1
## Performing GWAS evaluation

ms <- as.data.frame(t(mix1$scores))
ms$Locus <- rownames(ms)
MP2 <- merge(MP,ms,by="Locus",all.x = TRUE);
manhattan(MP2, pch=20,cex=.5, PVCN = "color score")

Be aware that the marker matrix M has to be imputed (no missing data allowed) and make sure that the number of rows in the M matrix is equivalent to the levels of the gTerm specified (i.e. if the gTerm is “id” and has 300 levels or in other words 300 individuals, then M has dimensions 300 x m, being m the number of markers).

Overlayed models [the overlay() function]

Another very useful function is the overlay function, which allows to overlay matrices of different random effects and estimate a single variance component for the overlayed terms.

data("DT_halfdiallel")
head(DT)

##   rep geno male female     sugar
## 1   1   12    1      2 13.950509
## 2   2   12    1      2  9.756918
## 3   1   13    1      3 13.906355
## 4   2   13    1      3  9.119455
## 5   1   14    1      4  5.174483
## 6   2   14    1      4  8.452221

DT$femalef <- as.factor(DT$female)
DT$malef <- as.factor(DT$male)
DT$genof <- as.factor(DT$geno)
#### model using overlay
modh <- mmer(sugar~1, 
             random=~vs(overlay(femalef,malef)) 
             + genof,
             data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -10.425   11:6:44      0           0
##     2      -6.487   11:6:44      0           0
##     3      -5.732   11:6:44      0           0
##     4      -5.67494   11:6:44      0           0
##     5      -5.67441   11:6:44      0           0

here the femalef and malef random effects are overlayed becoming a single random effect that has the same variance component.

Spatial models (using the 2-dimensional spline)

We will use the CPdata to show the use of 2-dimensional splines for accomodating spatial effects in field experiments. In early generation variety trials the availability of seed is low, which makes the use of unreplicated design a neccesity more than anything else. Experimental designs such as augmented designs and partially-replicated (p-rep) designs become every day more common this days.

In order to do a good job modeling the spatial trends happening in the field special covariance structures have been proposed to accomodate such spatial trends (i.e. autoregressive residuals; ar1). Unfortunately, some of these covariance structures make the modeling rather unstable. More recently other research groups have proposed the use of 2-dimensional splines to overcome such issues and have a more robust modeling of the spatial terms (Lee et al. 2013; Rodríguez-Álvarez et al. 2018).

In this example we assume an unreplicated population where row and range information is available which allows us to fit a 2 dimensional spline model.

data("DT_cpdata")
### mimic two fields
A <- A.mat(GT)
mix <- mmer(Yield~1,
            random=~vs(id, Gu=A) +
              vs(Rowf) +
              vs(Colf) +
              vs(spl2D(Row,Col)),
            rcov=~vs(units),
            data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -154.198   11:6:45      0           0
##     2      -152.064   11:6:46      1           0
##     3      -151.265   11:6:46      1           0
##     4      -151.202   11:6:47      2           0
##     5      -151.201   11:6:47      2           0

summary(mix)

## ============================================================
##          Multivariate Linear Mixed Model fit by REML         
## **********************  sommer 4.0  ********************** 
## ============================================================
##          logLik      AIC      BIC Method Converge
## Value -151.2011 304.4021 308.2938     NR     TRUE
## ============================================================
## Variance-Covariance components:
##                     VarComp VarCompSE Zratio Constraint
## u:id.Yield-Yield      783.4     319.3 2.4536   Positive
## u:Rowf.Yield-Yield    814.7     390.5 2.0863   Positive
## u:Colf.Yield-Yield    182.2     129.7 1.4053   Positive
## u:Row.Yield-Yield     513.6     694.7 0.7393   Positive
## u:units.Yield-Yield  2922.6     294.1 9.9368   Positive
## ============================================================
## Fixed effects:
##   Trait      Effect Estimate Std.Error t.value
## 1 Yield (Intercept)    132.1     8.791   15.03
## ============================================================
## Groups and observations:
##        Yield
## u:id     363
## u:Rowf    13
## u:Colf    36
## u:Row    168
## ============================================================
## Use the '$' sign to access results and parameters

Notice that the job is done by the spl2D() function that takes the Row and Col information to fit a spatial kernel.

Customized random effects

One of the most powerful features of sommer is the ability to provide any customized matrix and estimate any random effect. For example:

data(DT_cpdata)
GT[1:4,1:4]

##      scaffold_50439_2381 scaffold_39344_153 uneak_3436043 uneak_2632033
## P003                   0                  0             0             1
## P004                   0                  0             0             1
## P005                   0                 -1             0             1
## P006                  -1                 -1            -1             0

#### look at the data and fit the model
mix1 <- mmer(Yield~1,
              random=~vs(list(GT)),
              rcov=~units,
              data=DT)

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -286.365   11:6:48      1           0
##     2      -236.78   11:6:48      1           0
##     3      -200.635   11:6:49      2           0
##     4      -180.045   11:6:49      2           0
##     5      -176.4   11:6:49      2           0
##     6      -176.211   11:6:49      2           0
##     7      -176.207   11:6:50      3           0
##     8      -176.207   11:6:50      3           0

the matrix GT is provided as a random effect by encapsulating the matrix in a list and provided in the vs() function.

10) Genomic selection

In this section I decided to show the way you can fit an rrBLUP and GBLUP model in sommer using some wheat example data from CIMMYT in the genomic selection framework. This is the case of prediction of specific individuals within a population. It basically uses a similar model of the form:

\(y = X\beta + Zu + \epsilon\)

and takes advantage of the variance covariance matrix for the genotype effect known as the additive relationship matrix (A) and calculated using the A.mat function to establish connections among all individuals and predict the BLUPs for individuals that were not measured. In case the interest is to get BLUPs for markers the random effect is the actual marker matrix and the relationship among markers can be specified as well but in this example is assume a diagonal.

data("DT_wheat"); 
colnames(DT) <- paste0("X",1:ncol(DT))
DT <- as.data.frame(DT);DT$id <- as.factor(rownames(DT))
# select environment 1
rownames(GT) <- rownames(DT)
K <- A.mat(GT) # additive relationship matrix
colnames(K) <- rownames(K) <- rownames(DT)
# GBLUP pedigree-based approach
set.seed(12345)
y.trn <- DT
vv <- sample(rownames(DT),round(nrow(DT)/5))
y.trn[vv,"X1"] <- NA
head(y.trn)

##              X1          X2          X3         X4   id
## 775          NA -1.72746986 -1.89028479  0.0509159  775
## 2166 -0.2527028  0.40952243  0.30938553 -1.7387588 2166
## 2167  0.3418151 -0.64862633 -0.79955921 -1.0535691 2167
## 2465         NA  0.09394919  0.57046773  0.5517574 2465
## 3881         NA -0.28248062  1.61868192 -0.1142848 3881
## 3889  2.3360969  0.62647587  0.07353311  0.7195856 3889

## GBLUP
ans <- mmer(X1~1,
            random=~vs(id,Gu=K), 
            rcov=~units, 
            data=y.trn) # kinship based

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -202.344   11:6:59      1           0
##     2      -198.717   11:6:59      1           0
##     3      -197.634   11:7:0      2           0
##     4      -197.51   11:7:0      2           0
##     5      -197.508   11:7:1      3           0
##     6      -197.508   11:7:1      3           0

ans$U$`u:id`$X1 <- as.data.frame(ans$U$`u:id`$X1)
rownames(ans$U$`u:id`$X1) <- gsub("id","",rownames(ans$U$`u:id`$X1))
cor(ans$U$`u:id`$X1[vv,],DT[vv,"X1"], use="complete")

## [1] 0.4885674

## rrBLUP
ans2 <- mmer(X1~1,
             random=~vs(list(GT)), 
             rcov=~units,
             data=y.trn) # kinship based

## iteration    LogLik     wall    cpu(sec)   restrained
##     1      -343.082   11:7:3      1           0
##     2      -243.965   11:7:4      2           0
##     3      -208.257   11:7:5      3           0
##     4      -197.982   11:7:5      3           0
##     5      -197.519   11:7:6      4           0
##     6      -197.508   11:7:6      4           0
##     7      -197.508   11:7:7      5           0

u <- GT %*% as.matrix(ans2$U$`u:GT`$X1) # BLUPs for individuals
rownames(u) <- rownames(GT)
cor(u[vv,],DT[vv,"X1"]) # same correlation

## [1] 0.4885716

# the same can be applied in multi-response models in GBLUP or rrBLUP

11) Final remarks

Keep in mind that sommer uses direct inversion (DI) algorithm which can be very slow for large datasets. The package is focused in problems of the type p > n (more random effect levels than observations) and models with dense covariance structures. For example, for experiment with dense covariance structures with low-replication (i.e. 2000 records from 1000 individuals replicated twice with a covariance structure of 1000x1000) sommer will be faster than MME-based software. Also for genomic problems with large number of random effect levels, i.e. 300 individuals (n) with 100,000 genetic markers (p). For highly replicated trials with small covariance structures or n > p (i.e. 2000 records from 200 individuals replicated 10 times with covariance structure of 200x200) asreml or other MME-based algorithms will be much faster and we recommend you to opt for those software.

Literature

Covarrubias-Pazaran G. 2016. Genome assisted prediction of quantitative traits using the R package sommer. PLoS ONE 11(6):1-15.

Covarrubias-Pazaran G. 2018. Software update: Moving the R package sommer to multivariate mixed models for genome-assisted prediction. doi: https://doi.org/10.1101/354639

Bernardo Rex. 2010. Breeding for quantitative traits in plants. Second edition. Stemma Press. 390 pp.

Gilmour et al. 1995. Average Information REML: An efficient algorithm for variance parameter estimation in linear mixed models. Biometrics 51(4):1440-1450.

Henderson C.R. 1975. Best Linear Unbiased Estimation and Prediction under a Selection Model. Biometrics vol. 31(2):423-447.

Kang et al. 2008. Efficient control of population structure in model organism association mapping. Genetics 178:1709-1723.

Lee, D.-J., Durban, M., and Eilers, P.H.C. (2013). Efficient two-dimensional smoothing with P-spline ANOVA mixed models and nested bases. Computational Statistics and Data Analysis, 61, 22 - 37.

Lee et al. 2015. MTG2: An efficient algorithm for multivariate linear mixed model analysis based on genomic information. Cold Spring Harbor. doi: http://dx.doi.org/10.1101/027201.

Maier et al. 2015. Joint analysis of psychiatric disorders increases accuracy of risk prediction for schizophrenia, bipolar disorder, and major depressive disorder. Am J Hum Genet; 96(2):283-294.

Rodriguez-Alvarez, Maria Xose, et al. Correcting for spatial heterogeneity in plant breeding experiments with P-splines. Spatial Statistics 23 (2018): 52-71.

Searle. 1993. Applying the EM algorithm to calculating ML and REML estimates of variance components. Paper invited for the 1993 American Statistical Association Meeting, San Francisco.

Yu et al. 2006. A unified mixed-model method for association mapping that accounts for multiple levels of relatedness. Genetics 38:203-208.

Abdollahi Arpanahi R, Morota G, Valente BD, Kranis A, Rosa GJM, Gianola D. 2015. Assessment of bagging GBLUP for whole genome prediction of broiler chicken traits. Journal of Animal Breeding and Genetics 132:218-228.

Tunnicliffe W. 1989. On the use of marginal likelihood in time series model estimation. JRSS 51(1):15-27.

Quick start for the sommer package

Giovanny Covarrubias-Pazaran

2019-07-02