Calculate eigengenes for new data

Calculates eigengenes for modules in new data. By default, eigengenes are calculated from scratch using PCA, in a similar manner to the moduleEigengenes function. The function also offers a projection approach, which functions in a similar fashion to the predict method of prcomp.

Usage

# S4 method for class 'ModularExperiment,matrix'
calcEigengenes(
  object,
  newdata,
  project = FALSE,
  scale_reduced = TRUE,
  return_loadings = FALSE,
  scale_newdata = NULL,
  center_newdata = NULL,
  realign = TRUE,
  min_module_genes = 10
)

# S4 method for class 'ModularExperiment,data.frame'
calcEigengenes(
  object,
  newdata,
  project = FALSE,
  scale_reduced = TRUE,
  return_loadings = FALSE,
  scale_newdata = NULL,
  center_newdata = NULL,
  realign = TRUE,
  min_module_genes = 10
)

# S4 method for class 'ModularExperiment,SummarizedExperiment'
calcEigengenes(
  object,
  newdata,
  project = FALSE,
  scale_reduced = TRUE,
  assay_name = "normal",
  scale_newdata = NULL,
  center_newdata = NULL,
  realign = TRUE,
  min_module_genes = 10
)

# S4 method for class 'ModularExperiment'
predict(object, newdata, ...)

Arguments

object: A ModularExperiment object. By default, the scale and center slots are used to apply the original transformation to the new data. The loadings slot of this class will be used if project is TRUE.
newdata: New data for eigengenes to be calculated in. Must be a data.frame or matrix with features as rows and samples as columns, or a SummarizedExperiment object. Assumes that the rows of newdata match those of the ModularExperiment object.
project: If FALSE (default), calculate eigengenes from scratch in the new dataset using an approach similar to moduleEigengenes (i.e., performing PCA for each module in newdata). If FALSE, perform projection, using PCA rotation matrix from the original data to calculate module eigengenes. Projection approach is experimental.
scale_reduced: Whether or not the reduced data should be scaled after calculation.
return_loadings: If TRUE, additionally returns the feature loadings for the eigengenes.
scale_newdata: Controls whether the newdata are scaled. If NULL, performs scaling based on the ModularExperiment object's scale slot. The value of this argument will be passed to the scale argument of scale.
center_newdata: Controls whether the newdata are centered If NULL, performs centering based on the ModularExperiment object's center slot. The value of this argument will be passed to the center argument of scale.
realign: If project is TRUE, this argument is ignored. Else, controls whether eigengenes are realigned after PCA is performed to ensure the resultant signatures are positively correlated with average expression of the module. Similar to the align argument of moduleEigengenes.
min_module_genes: If project is FALSE, this argument is ignores. Else, controls the minimum number of genes required in a module for projection. Projected eigengenes are not calculated for modules with sizes below this threshold.
assay_name: If a SummarizedExperiment object is passed as new data, this argument indicates which assay should be used for projection.
...: Additional arguments to be passed to calcEigengenes.

Value

If return_loadings is TRUE, returns a list with the "reduced" matrix and "loadings" vector (one value per feature). If FALSE, returns only the reduced matrix.

The reduced matrix has samples as rows and modules as columns. If newdata was a matrix or data.frame, this will be returned as a matrix. If a SummarizedExperiment object was passed instead, then a If a ModularExperiment object will be created containing this matrix in its reduced slot.

Details

If scale_newdata and center_newdata are left as NULL, then the projection method assumes that the newdata are on the same scale as the original data of the object. It will therefore use the values of the center and scale slots of the object. For instance, if the scale slot is TRUE, the newdata will be scaled. If the scale slot is a vector, the values of this vector will be applied to scale the newdata.

Author

Jack Gisby

Examples

# Create ModularExperiment with random data (100 features, 50 samples,
# 10 modules)
me_1 <- ReducedExperiment:::.createRandomisedModularExperiment(100, 50, 10)

# Generate a new dataset with the same features (100 rows) but different
# samples/observations (20 columns)
X_2 <- ReducedExperiment:::.makeRandomData(100, 20, "gene", "sample")

# We can use the projection approach to calculate the eigengenes for
# the modules identified in dataset 1 for the samples in dataset 2
# This approach is based on the module loadings
me_2_project <- calcEigengenes(me_1, X_2, project = TRUE)
me_2_project[1:5, 1:5]
#>             module_3   module_4   module_5   module_7   module_9
#> sample_1  1.22860282  0.1098998 -0.3994967  0.6225404 -0.3399150
#> sample_2  0.03840139  1.4757127  1.5290504  2.3318725  0.6358647
#> sample_3  0.06706610 -1.2801196 -0.8433581 -0.5570758 -2.3058230
#> sample_4 -1.03371358 -0.2736773  0.7469042  0.4239787  0.7531614
#> sample_5 -0.05816772 -2.2594398  0.9274514  0.6076134 -0.8240435

# Alternatively, we can calculate eigengenes from scratch in the second
# dataset. This still uses the modules identified in the first dataset (me_1)
# but does not make use of the loadings. This approach is similar to
# that applied by WGCNA::moduleEigengenes.
me_2_eig <- calcEigengenes(me_1, X_2, project = FALSE)
me_2_eig[1:5, 1:5]
#>             module_1    module_2    module_3    module_4    module_5
#> sample_1 -0.77264363 -0.79497888 -0.03168875 -1.70172233  0.12220734
#> sample_2 -1.14156851  0.70822928  0.05088189 -0.17936889  2.64330795
#> sample_3  2.30464866  1.49382155  0.40186869 -0.01352815 -0.52443628
#> sample_4  0.09542815 -0.08564603  0.14493088  0.30611324  0.46941327
#> sample_5 -0.04783814 -1.62024124  0.95875606 -0.09346481 -0.04152795