DaMiR.normalization: Filter non Expressed and 'Hypervariant' features and Data...
In BioinfoMonzino/DaMiRseq: Data Mining for RNA-seq data: normalization, feature selection and classification

Description Usage Arguments Details Value Author(s) References See Also Examples

Features will be firstly filtered based on their expression value and/or by their variability across samples; features will be then normalized.

DaMiR.normalization(
  data,
  minCounts = 10,
  fSample = 0.5,
  hyper = c("yes", "no"),
  th.cv = 3,
  type = c("vst", "rlog", "logcpm"),
  nFitType = c("parametric", "local", "mean")
)

`data`	A `SummarizedExperiment` object
`minCounts`	Minimum reads counts; default is 10
`fSample`	Fraction of samples with `minCounts` counts; default is 0.5
`hyper`	Flag to enable gene filtering by Coefficient of Variation (CV); default is "yes"
`th.cv`	Threshold of minimum CV to consider a feature 'Hypervariant' accross samples; default is 3
`type`	Type of normalization to be applied: `varianceStabilizingTransformation` (`vst`), `rlog` or `logcpm` are allowed; default is "`vst`"
`nFitType`	Type of method to estimate the dispersion by vst or rlog. Default is "parametric".

Before normalization step, this function allows the user to filter features by:

Expression - Features will be filtered out whether their reads count do not reach a minCounts in at least fSample of samples;
CV - The CV of each feature is individually calculated for each sample class. Featurers with both class CV greater than th.cv will be discarded. Computing a class restricted CV may prevent the removal of hypervariant features that may be specifically associated with a certain class. This could be important, for example, for immune genes whose expression under definite conditions may unveil peculiar class-gene association.

Finally, expressed features will be normalized by varianceStabilizingTransformation (default) or rlog, both implemented in DESeq2 package. We suggest to use varianceStabilizingTransformation to speed up the normalization process because rlog is very time-consuming despite the two methods produce quite similar results.

A SummarizedExperiment object which contains a normalized expression matrix (log2 scale) and the data frame with 'class' and (optionally) variables.

Mattia Chiesa, Luca Piacentini

Michael I Love, Wolfgang Huber and Simon Anders (2014): Moderated estimation of fold change and dispersion for RNA-Seq data with DESeq2. Genome Biology

varianceStabilizingTransformation, rlog

# use example data:
data(SE)
# perform normalization on a subset of data:
SE_sub<-SE[1:1000, c(1:3, 21:23)]
data_norm <- DaMiR.normalization(SE_sub, minCounts=10, fSample=0.8,
hyper="yes", th.cv = 2.5)