MetabolomicsBasics (version 1.4.5)

ReplaceMissingValues: ReplaceMissingValues.

Description

ReplaceMissingValues will replace missing values within a numeric matrix based on a principal component analysis.

Usage

ReplaceMissingValues(x, ncomp = 10, silent = FALSE)

Value

A matrix of similar dimensions as x without missing values.

Arguments

x

Numeric matrix.

ncomp

Number of components to be used.

silent

FALSE, suppress messages setting silent=TRUE.

Details

The `nipals` algorithm is used to basically perform a PCA on the sparse matrix. Missing values are imputed based on the major components observed. Please check also the `impute.nipals` function from mixOmics which should basically give the same functionality since the 04/2021 update.

Examples

# load raw data and sample description
raw <- MetabolomicsBasics::raw
sam <- MetabolomicsBasics::sam

idx <- apply(raw, 2, CheckForOutliers, group = sam$GT, n_sd = 5, method = "logical")
sum(idx) # 215 values would be classified as outlier using a five-sigma band
old_vals <- raw[idx] # keep outlier values for comparison
raw_filt <- raw
raw_filt[idx] <- NA
raw_means <- apply(raw, 2, function(x) {
  sapply(split(x, sam$GT), mean, na.rm = TRUE)[as.numeric(sam$GT)]
})[idx]
raw_repl <- ReplaceMissingValues(x = raw_filt)
new_vals <- raw_repl[idx]
par(mfrow = c(2, 1))
breaks <- seq(-0.7, 1.3, 0.05)
hist(raw_means - old_vals, breaks = breaks, main = "", xlab = "Outliers", las = 1)
hist(raw_means - new_vals, breaks = breaks, main = "", xlab = "Replaced values", las = 1)