The function uses the data modelling output to generate corrected
data that can be used for downstream analysis. The input
is expected to be a SingleCellExperiment object that contains an
estimated ScpModel. There are two approaches:
scpKeepEffect(): keep the effects of interests. The
reconstructed data is the sum of the effect matrices for the
variable of interest and the residuals. Note that the intercepts
(baseline intensity of each feature) are not included by
default, but they can be added when intercept = TRUE.
scpRemoveBatchEffect(): remove any undesired effect. The batch
corrected data is the input data minus the effect matrices that
correspond to batch effect variables. Note that the intercepts
(baseline intensity of each feature) are removed by default, but
they can be kept when intercept = FALSE.
Despite the two approaches are conceptually different, they can lead to similar results if the effects that are used to reconstruct the data are the ones that are not removed when performing batch correction (see examples).
The function returns a new SingleCellExperiment that contains an
assay with the batch corrected data. Note that the 'ScpModel` is
erased in this new object.
scpKeepEffect(object, effects = NULL, intercept = FALSE, name)
scpRemoveBatchEffect(object, effects = NULL, intercept = TRUE, name)An object that inherits from the
SingleCellExperiment class. It must contain an estimated
ScpModel in its metadata
A character() vector. For scpKeepEffect(),
which model variable should be used to reconstruct the data.
For scpRemoveBatchEffect(), which model variable should be
removed from the data.
A logical(1). For scpKeepEffect(),
should the intercepts be included when reconstructing the
data? Defaults to FALSE, hence the intercepts are not
included. For scpRemoveBatchEffect(), should the intercepts
be removed from the data? Defaults to TRUE, hence the
intercepts are removed from the data.
A character(1) providing the name to use to retrieve
the model results. When retrieving a model and name is
missing, the name of the first model found in object is used.
ScpModel for functions to extract information from the
ScpModel object
ScpModel-Workflow to run a model on SCP data required for batch correction.
data("leduc_minimal")
scpModelFormula(leduc_minimal)
#> ~1 + Channel + Set + MedianIntensity + SampleType
reconstructed <- scpKeepEffect(leduc_minimal, effects = "SampleType")
batchCorreced <- scpRemoveBatchEffect(
leduc_minimal, effects = c("Channel", "Set", "MedianIntensity")
)
## The two approaches are identical
identical(reconstructed, batchCorreced)
#> [1] TRUE