diff --git a/R/Lrner.R b/R/Lrner.R
index bb59e62..75b1cba 100644
--- a/R/Lrner.R
+++ b/R/Lrner.R
@@ -28,15 +28,15 @@ Lrner <- R6Class("Lrner",
                    #' Learn parameters.
                    #' @param train_layer (`TrainLayer(1)`) \cr
                    #' Layer on which the learner is stored.
-                   #' @param na_rm (`logical(1)`) \cr
-                   #' If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.
+                   #' @param na_action `character(1)`\cr
+                   #' Handling of missing values. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" (only applicable on meta-data) to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided learner can handle missing values.
                    initialize = function (id,
                                           package = NULL,
                                           lrn_fct,
                                           param_train_list,
                                           param_pred_list = list(),
                                           train_layer,
-                                          na_rm = TRUE) {
+                                          na_action = "na.rm") {
                      private$id = id
                      private$package = package
                      private$lrn_fct = lrn_fct
@@ -50,11 +50,34 @@ Lrner <- R6Class("Lrner",
                      if (!any(c("TrainLayer", "TrainMetaLayer") %in% class(train_layer))) {
                        stop("A Lrner can only belong to a TrainLayer or a TrainMetaLayer object.")
                      }
-                     if (!is.logical(na_rm)) {
-                       stop("na.rm must be a logical value\n")
+                     # if (!is.logical(na_rm)) {
+                     #   stop("na.rm must be a logical value\n")
+                     # } else {
+                     #   private$na_rm = na_rm
+                     # }
+                     impute = FALSE
+                     if (na_action == "na.keep") {
+                       na_rm = FALSE
                      } else {
-                       private$na_rm = na_rm
+                       if (na_action == "na.rm") {
+                         na_rm = TRUE
+                       } else {
+                         if (na_action == "na.impute") {
+                           if ("TrainLayer" %in% class(train_layer)) {
+                             stop("Imputation is not yet handled for data modalities. Please use either the 'na.keep' or the 'na.rm' option.")
+                           }
+                           na_rm = FALSE
+                           impute = TRUE
+                           # Imputation takes place in the Training class, since it
+                           # happens after meta-data have been generated.
+                           train_layer$getTraining()$setImpute(impute = impute)
+                         } else {
+                           stop("na_action must be one of 'na.fails', 'na.rm' or 'na.impute'.")
+                         }
+                       }
                      }
+                     private$na_rm = na_rm
+                     # Instantiate a Lrner object
                      # Remove learner if already existing
                      if (train_layer$checkLrnerExist()) {
                        key_class = train_layer$getKeyClass()
@@ -135,7 +158,7 @@ Lrner <- R6Class("Lrner",
                        }
                      }
                      param_interface = data.frame(standard = c("x_name", "y_name", "object_name", "data_name"),
-                                               original = c(x, y, object, data))
+                                                  original = c(x, y, object, data))
                      private$param_interface = param_interface
                      private$extract_pred_fct = extract_pred_fct
                    },
diff --git a/R/TrainMetaLayer.R b/R/TrainMetaLayer.R
index ec36a69..eef1f12 100644
--- a/R/TrainMetaLayer.R
+++ b/R/TrainMetaLayer.R
@@ -157,7 +157,6 @@ TrainMetaLayer <- R6Class("TrainMetaLayer",
                             #' The list of parameters to call the imputation function. Not yet implemented!
                             #' @return
                             #' A new object with the predicted values is returned.
-                            #' @export
                             #'
                             impute = function (impute_fct = NULL,
                                                impute_param = NULL) {
diff --git a/R/VarSel.R b/R/VarSel.R
index 9a0eeda..70965f8 100644
--- a/R/VarSel.R
+++ b/R/VarSel.R
@@ -26,14 +26,15 @@ VarSel <- R6Class("VarSel",
                     #' Layer on which the learner is stored.
                     #' @param train_layer (`TrainLayer(1)`) \cr
                     #'  The training layer where to store the learner.
-                    #' @param na_rm (`logical(1)`) \cr
+                    #' @param na_action `character(1)`\cr
+                    #' Handling of missing values in meta-data. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" (only applicable on meta-data) to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided learner can handle missing values.
                     #' If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.
                     initialize = function (id,
                                            package = NULL,
                                            varsel_fct,
                                            varsel_param,
                                            train_layer,
-                                           na_rm = TRUE) {
+                                           na_action = "na.rm") {
                       private$id = id
                       private$package = package
                       private$varsel_fct = varsel_fct
@@ -46,11 +47,26 @@ VarSel <- R6Class("VarSel",
                       if (!any(c("TrainLayer") %in% class(train_layer))) {
                         stop("A variable selection tool can only belong to object of class TrainLayer.")
                       }
-                      if (!is.logical(na_rm)) {
-                        stop("na.rm must be a logical value\n")
+                      # if (!is.logical(na_rm)) {
+                      #   stop("na.rm must be a logical value\n")
+                      # } else {
+                      #   private$na_rm = na_rm
+                      # }
+
+                      if (na_action == "na.keep") {
+                        na_rm = FALSE
                       } else {
-                        private$na_rm = na_rm
+                        if (na_action == "na.rm") {
+                          na_rm = TRUE
+                        } else {
+                          if (na_action == "na.impute") {
+                              stop("Imputation is not yet handled for data modalities. Please use either the 'na.keep' or the 'na.rm' option.")
+                          } else {
+                            stop("na_action must be one of 'na.fails' or 'na.rm'.")
+                          }
+                        }
                       }
+                      private$na_rm = na_rm
                       # Remove VarSel if already existing
                       if (train_layer$checkVarSelExist()) {
                           key_class = train_layer$getKeyClass()
diff --git a/R/trainingFunctions.R b/R/trainingFunctions.R
index 9a472b9..b83101d 100644
--- a/R/trainingFunctions.R
+++ b/R/trainingFunctions.R
@@ -2,24 +2,25 @@
 #' @description
 #' Creates a [Training] object.
 #'
-#' @param id (`character(1)`) \cr
-#' @param ind_col (`character(1)`) \cr
-#' Name of column of individuals IDS.
-#' @param target (`character(1)`) \cr
+#' @param id `character` \cr
+#' Training's ID.
+#' @param target_df `data.frame` \cr
+#' Observed target values. A data frame with two columns: individual IDs and response variable values.
+#' @param ind_col `character` \cr
+#' Name of column of individuals IDs.
+#' @param target `character` \cr
 #' Name of the target variable.
-#' @param target_df (`data.frame(1)`) \cr
-#' Data frame with two columns: individual IDs and response variable values.
-#' @param problem_type (`character`) \cr
+#' @param problem_type `character` \cr
 #' Either "classification" or "regression".
-#' @param verbose (`boolean`) \cr
+#' @param verbose `boolean` \cr
 #' Warning messages will be displayed if set to TRUE.
 #' @return
 #' The created [Training] object is returned.
 #' @export
 createTraining = function (id,
+                           target_df,
                            ind_col,
                            target,
-                           target_df,
                            problem_type = "classification",
                            verbose = TRUE) {
   training = Training$new(
@@ -37,45 +38,45 @@ createTraining = function (id,
 #' @description
 #' Creates and store a [TrainLayer] on the [Training] object passed as argument.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training` \cr
 #' Training object where the created layer will be stored.
-#' @param train_layer_id (`character(1)`) \cr
+#' @param train_layer_id `character` \cr
 #' ID of the [TrainLayer] to be created.
-#' @param train_data (`data.frame(1)`) \cr
+#' @param train_data `data.frame(1)` \cr
 #' Data modality to be stored in [TrainData].
-#' @param varsel_package (`character(1)`) \cr
+#' @param varsel_package `character(1)` \cr
 #' Name of the package containing the function that implements the variable selection algorithm.\cr
-#' @param varsel_fct (`character(1)`) \cr
+#' @param varsel_fct `character(1)` \cr
 #' Name of the function that performs variable selection. For the default value NULL no variable selection will be performed.
-#' @param varsel_param (`list`) \cr
+#' @param varsel_param `list` \cr
 #' List of arguments to be passed to \code{varsel_fct}.
-#' @param lrner_package (`character(1)`) \cr
+#' @param lrner_package `character(1)` \cr
 #' Name of the package containing the function that implements the learning algorithm.\cr
-#' @param lrn_fct  `(character(1)` \cr
+#' @param lrn_fct  `character(1)` \cr
 #' Name of the function that that implements the learning algorithm.
-#' @param param_train_list  `(character(1)` \cr
+#' @param param_train_list  `character(1)` \cr
 #' List of arguments to be passed to \code{lrn_fct}.
-#' @param param_pred_list   `(character(1)` \cr
+#' @param param_pred_list   `character(1)` \cr
 #' List of arguments to be passed to \code{predict} when computing predictions.
-#' @param na_rm \cr
-#' If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.
-#' @param x (`character(1)`) \cr
+#' @param na_action `character(1)`\cr
+#' Handling of missing values in data a modality. Set to "na.keep" to keep missing values, or "na.rm" to remove individuals with missing values. Imputation of missing values in data modalities ist not yet handled.
+#' @param x `character(1)` \cr
 #' If the name of the argument used by the provided original functions to pass
 #' the matrix of independent variable is not \code{x}, use this argument to specify how it is callled in the provided function.
-#' @param y (`character(1)`) \cr
+#' @param y `character(1)` \cr
 #' If the name of the argument used by the provided original functions to pass
 #' the target variable is not \code{y}, use this argument to specify how it is callled in the provided function.
-#' @param object (`character(1)`) \cr
+#' @param object `character(1)` \cr
 #' The generic function \code{predict} uses a parameter \code{object} to pass a model.
 #' If the corresponding argument is named differently in your predict function, specify the name.
-#' @param data (`character(1)`) \cr
+#' @param data `character(1)` \cr
 #' The generic function \code{predict} uses a parameter \code{data} to pass new data.
 #' If the corresponding argument is named differently in your predict function, specify the name.
-#' @param extract_pred_fct (`character(1) or function(1)`) \cr
+#' @param extract_pred_fct `character(1) or function(1)` \cr
 #' If the predict function that is called for the model does not return a vector, then
 #' use this argument to specify a (or a name of a) function that can be used to extract vector of predictions.
 #' Default value is NULL, if predictions are in a column.
-#' @param extract_var_fct (`character(1) or function(1)`) \cr
+#' @param extract_var_fct `character(1) or function(1)` \cr
 #' If the variable selection function that is called does not return a vector, then
 #' use this argument to specify a (or a name of a) function that can be used to extract vector of selected variables.
 #' Default value is NULL, if selected variables are in a vector.
@@ -93,7 +94,7 @@ createTrainLayer = function (training,
                              lrn_fct,
                              param_train_list = list(),
                              param_pred_list = list(),
-                             na_rm = TRUE,
+                             na_action = "na.rm",
                              x = "x",
                              y = "y",
                              object = "object",
@@ -115,7 +116,7 @@ createTrainLayer = function (training,
       varsel_fct = varsel_fct,
       varsel_param = varsel_param,
       train_layer = train_layer,
-      na_rm = na_rm
+      na_action = na_action
     )
     var_sel$interface(x = x,
                       y = y,
@@ -131,7 +132,7 @@ createTrainLayer = function (training,
     param_train_list = param_train_list,
     param_pred_list = param_pred_list,
     train_layer = train_layer,
-    na_rm = na_rm
+    na_action = na_action
   )
   lrner$interface(x = x,
                   y = y,
@@ -151,33 +152,33 @@ createTrainLayer = function (training,
 #' @description
 #' Creates and store a [TrainMetaLayer] on the [Training] object passed as argument.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training(1)` \cr
 #' Training object where the created layer will be stored.
-#' @param meta_layer_id (`character(1)`) \cr
+#' @param meta_layer_id `character(1)` \cr
 #' ID of the layer to be created.
-#' @param lrner_package (`character(1)`) \cr
+#' @param lrner_package `character(1)` \cr
 #' Name of the package containing the function that implements the learning algorithm.\cr
-#' @param lrn_fct  `(character(1)` \cr
+#' @param lrn_fct  `character(1)` \cr
 #' Name of the function that that implements the learning algorithm.
-#' @param param_train_list  `(character(1)` \cr
+#' @param param_train_list  `character(1)` \cr
 #' List of arguments to be passed to \code{lrn_fct}.
-#' @param param_pred_list   `(character(1)` \cr
+#' @param param_pred_list   `list(1)` \cr
 #' List of arguments to be passed to \code{predict} when computing predictions.
-#' @param na_action \cr
+#' @param na_action `character(1)`\cr
 #' Handling of missing values in meta-data. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided meta-learner can handle missing values.
-#' @param x (`character(1)`) \cr
+#' @param x `character(1)` \cr
 #' If the name of the argument used by the provided original functions to pass
 #' the matrix of independent variable is not \code{x}, use this argument to specify how it is callled in the provided function.
-#' @param y (`character(1)`) \cr
+#' @param y `character(1)` \cr
 #' If the name of the argument used by the provided original functions to pass
 #' the target variable is not \code{y}, use this argument to specify how it is callled in the provided function.
-#' @param object (`character(1)`) \cr
+#' @param object `character(1)` \cr
 #' The generic function \code{predict} uses a parameter \code{object} to pass a model.
 #' If the corresponding argument is named differently in your predict function, specify the name.
-#' @param data (`character(1)`) \cr
+#' @param data `character(1)` \cr
 #' The generic function \code{predict} uses a parameter \code{data} to pass new data.
 #' If the corresponding argument is named differently in your predict function, specify the name.
-#' @param extract_pred_fct (`character(1) or function(1)`) \cr
+#' @param extract_pred_fct `character(1) or function(1)` \cr
 #' If the predict function that is called for the model does not return a vector, then
 #' use this argument to specify a (or a name of a) function that can be used to extract vector of predictions.
 #' Default value is NULL, if predictions are in a vector.
@@ -200,23 +201,6 @@ createTrainMetaLayer = function (training,
   # Instantiate a layer
   train_meta_layer = TrainMetaLayer$new(id = meta_layer_id,
                                         training = training)
-  impute = FALSE
-  # Instantiate a Lrner object
-  if (na_action == "na.keep") {
-    na_rm = FALSE
-  } else {
-    if (na_action == "na.rm") {
-      na_rm = TRUE
-    } else {
-      if (na_action == "na.impute") {
-        na_rm = FALSE
-        impute = TRUE
-      } else {
-        stop("na_action must be one of 'na.fails', 'na.rm' or 'na.impute'.")
-      }
-    }
-  }
-  training$setImpute(impute = impute)
   meta_lrner = Lrner$new(
     id = sprintf("%s_lrner", meta_layer_id),
     package = lrner_package,
@@ -224,7 +208,7 @@ createTrainMetaLayer = function (training,
     param_train_list = param_train_list,
     param_pred_list = param_pred_list,
     train_layer = train_meta_layer,
-    na_rm = na_rm
+    na_action = na_action
   )
   meta_lrner$interface(x = x,
                        y = y,
@@ -238,7 +222,7 @@ createTrainMetaLayer = function (training,
 #' @description
 #' Variable selection on the current training object.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training(1)` \cr
 #' Training object where the created layer will be stored.
 #' @param ind_subset `vector(1)` \cr
 #' ID subset of individuals to be used for variable selection.
@@ -258,15 +242,15 @@ varSelection = function (training,
 #' @description
 #' Trains the [Training] object passed as argument. All leaners and the meta learner are trained.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training(1)` \cr
 #' Training object where the created layer will be stored.
-#' @param ind_subset (`vector(1)`) \cr
+#' @param ind_subset `vector(1)` \cr
 #' ID subset to be used for training.
 #' @param use_var_sel `boolean(1)` \cr
 #' If TRUE, variable selection is performed before training.
-#' @param resampling_method (`function(1)`) \cr
+#' @param resampling_method `function(1)` \cr
 #' Function for internal validation. If not specify, the \code{resampling} function from the package \code{caret} is used for a 10-folds cross-validation.
-#' @param resampling_arg (`list(1)`) \cr
+#' @param resampling_arg `list(1)` \cr
 #' List of arguments to be passed to the function.
 #'
 #' @return
@@ -291,11 +275,11 @@ fusemlr = function (training,
 #' @description
 #' Computes predictions for the [Testing] object passed as argument.
 #'
-#' @param object (`Training(1)`) \cr
+#' @param object `Training(1)` \cr
 #' Training object to be used to compute predictions.
-#' @param testing (`Testing(1)`) \cr
+#' @param testing `Testing(1)` \cr
 #' A new testing object to be predicted.
-#' @param ind_subset (`vector(1)`) \cr
+#' @param ind_subset `vector(1)` \cr
 #' Vector of IDs to be predicted.
 #'
 #' @return
@@ -314,7 +298,7 @@ predict.Training = function (object,
 #' @description
 #' Extracts models stored on each layers; base and meta models are extracted.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training(1)` \cr
 #' The [Training] object of interest.
 #'
 #' @return
@@ -329,7 +313,7 @@ extractModel = function (training) {
 #' @description
 #' Extracts data stored on each layers; base and meta data are extracted.
 #'
-#' @param training (`Training(1)`) \cr
+#' @param training `Training(1)` \cr
 #' The [Training] object of interest.
 #'
 #' @return
@@ -344,7 +328,7 @@ extractData = function (training) {
 #' @description
 #' Summaries a fuseMLR [Training] object.
 #'
-#' @param object (`Training(1)`) \cr
+#' @param object `Training(1)` \cr
 #' The [Training] object of interest.
 #' @param ... \cr
 #' Further arguments.
@@ -359,7 +343,7 @@ summary.Training = function (object, ...) {
 #' @description
 #' An upset plot of overlapping individuals.
 #'
-#' @param object (`Training(1) or Testing(1)`) \cr
+#' @param object `Training(1) or Testing(1)` \cr
 #' Training or testing object for each the upset plot will be created.
 #' @param ... \cr
 #' Further arguments to be passed to the \code{upset} function from package \code{UpSetR}.
diff --git a/README.Rmd b/README.Rmd
index ea6d034..fb8272b 100644
--- a/README.Rmd
+++ b/README.Rmd
@@ -103,7 +103,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(),
-                 na_rm = FALSE)
+                 na_action = "na.keep")
 
 # Create gene protein abundance layer
 createTrainLayer(training = training,
@@ -121,7 +121,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(),
-                 na_rm = FALSE)
+                 na_action = "na.keep")
 
 # Create methylation layer
 createTrainLayer(training = training,
@@ -139,7 +139,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(),
-                 na_rm = FALSE)
+                 na_action = "na.keep")
 ```
 
 - Also add a meta layer.
@@ -293,7 +293,7 @@ createTrainLayer(training = training,
                                          kernel = 'radial',
                                          probability = TRUE),
                  param_pred_list = list(probability = TRUE),
-                 na_rm = TRUE,
+                 na_action = "na.keep",
                  x = "x",
                  y = "y",
                  object = "object",
diff --git a/man/Lrner.Rd b/man/Lrner.Rd
index 00688e0..14f39d6 100644
--- a/man/Lrner.Rd
+++ b/man/Lrner.Rd
@@ -39,7 +39,7 @@ Initialize a default parameters list.
   param_train_list,
   param_pred_list = list(),
   train_layer,
-  na_rm = TRUE
+  na_action = "na.rm"
 )}\if{html}{\out{</div>}}
 }
 
@@ -65,8 +65,8 @@ Learn parameters.}
 \item{\code{train_layer}}{(\code{TrainLayer(1)}) \cr
 Layer on which the learner is stored.}
 
-\item{\code{na_rm}}{(\code{logical(1)}) \cr
-If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.}
+\item{\code{na_action}}{\code{character(1)}\cr
+Handling of missing values. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" (only applicable on meta-data) to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided learner can handle missing values.}
 }
 \if{html}{\out{</div>}}
 }
diff --git a/man/VarSel.Rd b/man/VarSel.Rd
index b9dbfdd..49cd251 100644
--- a/man/VarSel.Rd
+++ b/man/VarSel.Rd
@@ -37,7 +37,7 @@ Learner ID.
   varsel_fct,
   varsel_param,
   train_layer,
-  na_rm = TRUE
+  na_action = "na.rm"
 )}\if{html}{\out{</div>}}
 }
 
@@ -61,7 +61,8 @@ Layer on which the learner is stored.}
 \item{\code{train_layer}}{(\code{TrainLayer(1)}) \cr
 The training layer where to store the learner.}
 
-\item{\code{na_rm}}{(\code{logical(1)}) \cr
+\item{\code{na_action}}{\code{character(1)}\cr
+Handling of missing values in meta-data. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" (only applicable on meta-data) to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided learner can handle missing values.
 If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.}
 }
 \if{html}{\out{</div>}}
diff --git a/man/createTrainLayer.Rd b/man/createTrainLayer.Rd
index 21607d5..1ffd716 100644
--- a/man/createTrainLayer.Rd
+++ b/man/createTrainLayer.Rd
@@ -15,7 +15,7 @@ createTrainLayer(
   lrn_fct,
   param_train_list = list(),
   param_pred_list = list(),
-  na_rm = TRUE,
+  na_action = "na.rm",
   x = "x",
   y = "y",
   object = "object",
@@ -25,61 +25,61 @@ createTrainLayer(
 )
 }
 \arguments{
-\item{training}{(\code{Training(1)}) \cr
+\item{training}{\code{Training} \cr
 Training object where the created layer will be stored.}
 
-\item{train_layer_id}{(\code{character(1)}) \cr
+\item{train_layer_id}{\code{character} \cr
 ID of the \link{TrainLayer} to be created.}
 
-\item{train_data}{(\code{data.frame(1)}) \cr
+\item{train_data}{\code{data.frame(1)} \cr
 Data modality to be stored in \link{TrainData}.}
 
-\item{varsel_package}{(\code{character(1)}) \cr
+\item{varsel_package}{\code{character(1)} \cr
 Name of the package containing the function that implements the variable selection algorithm.\cr}
 
-\item{varsel_fct}{(\code{character(1)}) \cr
+\item{varsel_fct}{\code{character(1)} \cr
 Name of the function that performs variable selection. For the default value NULL no variable selection will be performed.}
 
-\item{varsel_param}{(\code{list}) \cr
+\item{varsel_param}{\code{list} \cr
 List of arguments to be passed to \code{varsel_fct}.}
 
-\item{lrner_package}{(\code{character(1)}) \cr
+\item{lrner_package}{\code{character(1)} \cr
 Name of the package containing the function that implements the learning algorithm.\cr}
 
-\item{lrn_fct}{\verb{(character(1)} \cr
+\item{lrn_fct}{\code{character(1)} \cr
 Name of the function that that implements the learning algorithm.}
 
-\item{param_train_list}{\verb{(character(1)} \cr
+\item{param_train_list}{\code{character(1)} \cr
 List of arguments to be passed to \code{lrn_fct}.}
 
-\item{param_pred_list}{\verb{(character(1)} \cr
+\item{param_pred_list}{\code{character(1)} \cr
 List of arguments to be passed to \code{predict} when computing predictions.}
 
-\item{na_rm}{\cr
-If \code{TRUE}, the individuals with missing predictor values will be removed from the training dataset.}
+\item{na_action}{\code{character(1)}\cr
+Handling of missing values in data a modality. Set to "na.keep" to keep missing values, or "na.rm" to remove individuals with missing values. Imputation of missing values in data modalities ist not yet handled.}
 
-\item{x}{(\code{character(1)}) \cr
+\item{x}{\code{character(1)} \cr
 If the name of the argument used by the provided original functions to pass
 the matrix of independent variable is not \code{x}, use this argument to specify how it is callled in the provided function.}
 
-\item{y}{(\code{character(1)}) \cr
+\item{y}{\code{character(1)} \cr
 If the name of the argument used by the provided original functions to pass
 the target variable is not \code{y}, use this argument to specify how it is callled in the provided function.}
 
-\item{object}{(\code{character(1)}) \cr
+\item{object}{\code{character(1)} \cr
 The generic function \code{predict} uses a parameter \code{object} to pass a model.
 If the corresponding argument is named differently in your predict function, specify the name.}
 
-\item{data}{(\code{character(1)}) \cr
+\item{data}{\code{character(1)} \cr
 The generic function \code{predict} uses a parameter \code{data} to pass new data.
 If the corresponding argument is named differently in your predict function, specify the name.}
 
-\item{extract_pred_fct}{(\verb{character(1) or function(1)}) \cr
+\item{extract_pred_fct}{\verb{character(1) or function(1)} \cr
 If the predict function that is called for the model does not return a vector, then
 use this argument to specify a (or a name of a) function that can be used to extract vector of predictions.
 Default value is NULL, if predictions are in a column.}
 
-\item{extract_var_fct}{(\verb{character(1) or function(1)}) \cr
+\item{extract_var_fct}{\verb{character(1) or function(1)} \cr
 If the variable selection function that is called does not return a vector, then
 use this argument to specify a (or a name of a) function that can be used to extract vector of selected variables.
 Default value is NULL, if selected variables are in a vector.}
diff --git a/man/createTrainMetaLayer.Rd b/man/createTrainMetaLayer.Rd
index b2bd331..3d6d980 100644
--- a/man/createTrainMetaLayer.Rd
+++ b/man/createTrainMetaLayer.Rd
@@ -20,44 +20,44 @@ createTrainMetaLayer(
 )
 }
 \arguments{
-\item{training}{(\code{Training(1)}) \cr
+\item{training}{\code{Training(1)} \cr
 Training object where the created layer will be stored.}
 
-\item{meta_layer_id}{(\code{character(1)}) \cr
+\item{meta_layer_id}{\code{character(1)} \cr
 ID of the layer to be created.}
 
-\item{lrner_package}{(\code{character(1)}) \cr
+\item{lrner_package}{\code{character(1)} \cr
 Name of the package containing the function that implements the learning algorithm.\cr}
 
-\item{lrn_fct}{\verb{(character(1)} \cr
+\item{lrn_fct}{\code{character(1)} \cr
 Name of the function that that implements the learning algorithm.}
 
-\item{param_train_list}{\verb{(character(1)} \cr
+\item{param_train_list}{\code{character(1)} \cr
 List of arguments to be passed to \code{lrn_fct}.}
 
-\item{param_pred_list}{\verb{(character(1)} \cr
+\item{param_pred_list}{\code{list(1)} \cr
 List of arguments to be passed to \code{predict} when computing predictions.}
 
-\item{na_action}{\cr
+\item{na_action}{\code{character(1)}\cr
 Handling of missing values in meta-data. Set to "na.keep" to keep missing values, "na.rm" to remove individuals with missing values or "na.impute" to impute missing values in meta-data. Only median and mode based imputations are actually handled. With the "na.keep" option, ensure that the provided meta-learner can handle missing values.}
 
-\item{x}{(\code{character(1)}) \cr
+\item{x}{\code{character(1)} \cr
 If the name of the argument used by the provided original functions to pass
 the matrix of independent variable is not \code{x}, use this argument to specify how it is callled in the provided function.}
 
-\item{y}{(\code{character(1)}) \cr
+\item{y}{\code{character(1)} \cr
 If the name of the argument used by the provided original functions to pass
 the target variable is not \code{y}, use this argument to specify how it is callled in the provided function.}
 
-\item{object}{(\code{character(1)}) \cr
+\item{object}{\code{character(1)} \cr
 The generic function \code{predict} uses a parameter \code{object} to pass a model.
 If the corresponding argument is named differently in your predict function, specify the name.}
 
-\item{data}{(\code{character(1)}) \cr
+\item{data}{\code{character(1)} \cr
 The generic function \code{predict} uses a parameter \code{data} to pass new data.
 If the corresponding argument is named differently in your predict function, specify the name.}
 
-\item{extract_pred_fct}{(\verb{character(1) or function(1)}) \cr
+\item{extract_pred_fct}{\verb{character(1) or function(1)} \cr
 If the predict function that is called for the model does not return a vector, then
 use this argument to specify a (or a name of a) function that can be used to extract vector of predictions.
 Default value is NULL, if predictions are in a vector.}
diff --git a/man/createTraining.Rd b/man/createTraining.Rd
index 53050db..277d068 100644
--- a/man/createTraining.Rd
+++ b/man/createTraining.Rd
@@ -6,29 +6,30 @@
 \usage{
 createTraining(
   id,
+  target_df,
   ind_col,
   target,
-  target_df,
   problem_type = "classification",
   verbose = TRUE
 )
 }
 \arguments{
-\item{id}{(\code{character(1)}) \cr}
+\item{id}{\code{character} \cr
+Training's ID.}
 
-\item{ind_col}{(\code{character(1)}) \cr
-Name of column of individuals IDS.}
+\item{target_df}{\code{data.frame} \cr
+Observed target values. A data frame with two columns: individual IDs and response variable values.}
 
-\item{target}{(\code{character(1)}) \cr
-Name of the target variable.}
+\item{ind_col}{\code{character} \cr
+Name of column of individuals IDs.}
 
-\item{target_df}{(\code{data.frame(1)}) \cr
-Data frame with two columns: individual IDs and response variable values.}
+\item{target}{\code{character} \cr
+Name of the target variable.}
 
-\item{problem_type}{(\code{character}) \cr
+\item{problem_type}{\code{character} \cr
 Either "classification" or "regression".}
 
-\item{verbose}{(\code{boolean}) \cr
+\item{verbose}{\code{boolean} \cr
 Warning messages will be displayed if set to TRUE.}
 }
 \value{
diff --git a/man/extractData.Rd b/man/extractData.Rd
index f695a37..2594c54 100644
--- a/man/extractData.Rd
+++ b/man/extractData.Rd
@@ -7,7 +7,7 @@
 extractData(training)
 }
 \arguments{
-\item{training}{(\code{Training(1)}) \cr
+\item{training}{\code{Training(1)} \cr
 The \link{Training} object of interest.}
 }
 \value{
diff --git a/man/fusemlr.Rd b/man/fusemlr.Rd
index c6af1f0..9a45651 100644
--- a/man/fusemlr.Rd
+++ b/man/fusemlr.Rd
@@ -13,19 +13,19 @@ fusemlr(
 )
 }
 \arguments{
-\item{training}{(\code{Training(1)}) \cr
+\item{training}{\code{Training(1)} \cr
 Training object where the created layer will be stored.}
 
-\item{ind_subset}{(\code{vector(1)}) \cr
+\item{ind_subset}{\code{vector(1)} \cr
 ID subset to be used for training.}
 
 \item{use_var_sel}{\code{boolean(1)} \cr
 If TRUE, variable selection is performed before training.}
 
-\item{resampling_method}{(\verb{function(1)}) \cr
+\item{resampling_method}{\verb{function(1)} \cr
 Function for internal validation. If not specify, the \code{resampling} function from the package \code{caret} is used for a 10-folds cross-validation.}
 
-\item{resampling_arg}{(\code{list(1)}) \cr
+\item{resampling_arg}{\code{list(1)} \cr
 List of arguments to be passed to the function.}
 }
 \value{
diff --git a/man/predict.Training.Rd b/man/predict.Training.Rd
index f3dece4..4b9473b 100644
--- a/man/predict.Training.Rd
+++ b/man/predict.Training.Rd
@@ -7,13 +7,13 @@
 \method{predict}{Training}(object, testing, ind_subset = NULL)
 }
 \arguments{
-\item{object}{(\code{Training(1)}) \cr
+\item{object}{\code{Training(1)} \cr
 Training object to be used to compute predictions.}
 
-\item{testing}{(\code{Testing(1)}) \cr
+\item{testing}{\code{Testing(1)} \cr
 A new testing object to be predicted.}
 
-\item{ind_subset}{(\code{vector(1)}) \cr
+\item{ind_subset}{\code{vector(1)} \cr
 Vector of IDs to be predicted.}
 }
 \value{
diff --git a/man/summary.Training.Rd b/man/summary.Training.Rd
index 384e04d..10cb353 100644
--- a/man/summary.Training.Rd
+++ b/man/summary.Training.Rd
@@ -7,7 +7,7 @@
 \method{summary}{Training}(object, ...)
 }
 \arguments{
-\item{object}{(\code{Training(1)}) \cr
+\item{object}{\code{Training(1)} \cr
 The \link{Training} object of interest.}
 
 \item{...}{\cr
diff --git a/man/upsetplot.Rd b/man/upsetplot.Rd
index 862fd2d..4c66ec4 100644
--- a/man/upsetplot.Rd
+++ b/man/upsetplot.Rd
@@ -7,7 +7,7 @@
 upsetplot(object, ...)
 }
 \arguments{
-\item{object}{(\verb{Training(1) or Testing(1)}) \cr
+\item{object}{\verb{Training(1) or Testing(1)} \cr
 Training or testing object for each the upset plot will be created.}
 
 \item{...}{\cr
diff --git a/man/varSelection.Rd b/man/varSelection.Rd
index e88f124..a1df063 100644
--- a/man/varSelection.Rd
+++ b/man/varSelection.Rd
@@ -7,7 +7,7 @@
 varSelection(training, ind_subset = NULL)
 }
 \arguments{
-\item{training}{(\code{Training(1)}) \cr
+\item{training}{\code{Training(1)} \cr
 Training object where the created layer will be stored.}
 
 \item{ind_subset}{\code{vector(1)} \cr
diff --git a/tests/testthat/test-Training.R b/tests/testthat/test-Training.R
index b3128e5..1791704 100644
--- a/tests/testthat/test-Training.R
+++ b/tests/testthat/test-Training.R
@@ -176,7 +176,7 @@ test_that("Training: all tests", {
     lrner_meta <- Lrner$new(id = "weighted",
                             lrn_fct = "weightedMeanLearner",
                             param_train_list = list(),
-                            na_rm = FALSE,
+                            na_action = "na.keep",
                             train_layer = tl_meta)
   })
 
diff --git a/vignettes/how_to_use.Rmd b/vignettes/how_to_use.Rmd
index f1c88ac..62792f4 100644
--- a/vignettes/how_to_use.Rmd
+++ b/vignettes/how_to_use.Rmd
@@ -77,7 +77,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(),
-                 na_rm = TRUE)
+                 na_action = "na.keep")
 
 # Create gene protein abundance layer
 createTrainLayer(training = training,
@@ -95,7 +95,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(type = "response"),
-                 na_rm = TRUE)
+                 na_action = "na.keep")
 
 # Create methylation layer
 createTrainLayer(training = training,
@@ -113,7 +113,7 @@ createTrainLayer(training = training,
                                          mtry = 1L,
                                      na.action = "na.learn"),
                  param_pred_list = list(),
-                 na_rm = TRUE)
+                 na_action = "na.keep")
 ```
 
 Also add a meta layer.
@@ -264,7 +264,7 @@ createTrainLayer(training = training,
                                          kernel = 'radial',
                                          probability = TRUE),
                  param_pred_list = list(probability = TRUE),
-                 na_rm = TRUE,
+                 na_action = "na.rm"),
                  x = "x",
                  y = "y",
                  object = "object",