Median normalization — med.norm • PRECISION.array

Normalize a training dataset so that each array shares a same median and store the median from the training dataset as the reference to frozen median normalize a test dataset. Also two other options are available: to only normalize a training dataset but not frozen normalize a test dataset, or vise versa.

Usage

med.norm(train = NULL, test = NULL, ref.dis = NULL)

Arguments

train: the training dataset to be median normalized. The dataset must have rows as probes and columns as samples. This can be left unspecified if ref.dis is suppied for frozen normalize test set.
test: the test dataset to be frozen median normalized. The dataset must have rows as probes and columns as samples. The number of rows must equal to the number of rows in the training set. By default, the test set is not specified (test = NULL) and no frozen normalization will be performed.
ref.dis: the reference distribution for frozen median normalize test set against previously normalized training set. This is required when train is not supplied. By default, ref.dis = NULL.

Value

a list of two datasets and one reference distribution:

train.mn: the normalized training set, if training set is specified
test.fmn: the frozen normalized test set, if test set is specified
ref.dis: the reference distribution

Examples

set.seed(101)
group.id <- substr(colnames(nuhdata.pl), 7, 7)
train.ind <- colnames(nuhdata.pl)[c(sample(which(group.id == "E"), size = 64),
                               sample(which(group.id == "V"), size = 64))]
train.dat <- nuhdata.pl[, train.ind]
test.dat <- nuhdata.pl[, !colnames(nuhdata.pl) %in% train.ind]

# normalize only training set
data.mn <- med.norm(train = train.dat)
str(data.mn)
#> List of 3
#>  $ train.mn: num [1:1810, 1:128] 8.5 8.39 8.14 8.47 8.45 ...
#>   ..- attr(*, "dimnames")=List of 2
#>   .. ..$ : chr [1:1810] "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" ...
#>   .. ..$ : chr [1:128] "GL5140E" "JB5556E" "JB4783E" "GL4527E" ...
#>  $ test.fmn: NULL
#>  $ ref.dis : num 5.43

# normalize training set and frozen normalize test set
data.mn <- med.norm(train = train.dat, test = test.dat)
str(data.mn)
#> List of 3
#>  $ train.mn: num [1:1810, 1:128] 8.5 8.39 8.14 8.47 8.45 ...
#>   ..- attr(*, "dimnames")=List of 2
#>   .. ..$ : chr [1:1810] "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" ...
#>   .. ..$ : chr [1:128] "GL5140E" "JB5556E" "JB4783E" "GL4527E" ...
#>  $ test.fmn: num [1:1810, 1:64] 6.48 6.92 7.21 5.92 6.14 ...
#>   ..- attr(*, "dimnames")=List of 2
#>   .. ..$ : chr [1:1810] "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" ...
#>   .. ..$ : chr [1:64] "JB4166E" "JB5669E" "JB4112E" "JB5847E" ...
#>  $ ref.dis : num 5.43

# frozen normalize test set with reference distribution
ref <- med.norm(train = train.dat)$ref.dis
data.mn <- med.norm(test = test.dat, ref.dis = ref)
str(data.mn)
#> List of 3
#>  $ train.mn: NULL
#>  $ test.fmn: num [1:1810, 1:64] 6.48 6.92 7.21 5.92 6.14 ...
#>   ..- attr(*, "dimnames")=List of 2
#>   .. ..$ : chr [1:1810] "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" "A_25_P00011991" ...
#>   .. ..$ : chr [1:64] "JB4166E" "JB5669E" "JB4112E" "JB5847E" ...
#>  $ ref.dis : num 5.43