# CheckForOutliers: CheckForOutliers. In MetabolomicsBasics: Basic Functions to Investigate Metabolomics Data Matrices

 CheckForOutliers R Documentation

## CheckForOutliers.

### Description

CheckForOutliers will evaluate a numeric vector and check if outliers within groups based on group mean \pm n \times sd.

### Usage

CheckForOutliers(
x = NULL,
group = NULL,
n_sd = 3,
method = c("idx", "logical", "dist")
)


### Arguments

 x Numeric vector. group Factor vector of length(x). n_sd Cutoff for outliers in E being mean(group)+-n_sd*sd(group) where group values are calculated without the outlier candidate. method Different variants of the result value. See details.

### Details

The numeric will be split by groups and each value will be evaluated with respect to its distance to the group mean (calculated out of the other values in the group). Distance here means the number of standard deviations the value is off the group mean. With different choices of method the output can be switched from the calculated fold-distances to a boolean of length(x) or and Index vector giving the outliers directly (see examples).

### Value

Depending on the selected method. See details.

### Examples

set.seed(0)
x <- runif(10)
x[1] <- 2
group <- gl(2, 5)
CheckForOutliers(x, group, method = "dist")
CheckForOutliers(x, group, method = "logical")
CheckForOutliers(x, group, method = "idx")
graphics::par(mfrow = c(1, 2))
bg <- c(3, 2)[1 + CheckForOutliers(x, group, method = "logical")]
graphics::plot(x = as.numeric(group), y = x, pch = 21, cex = 3,
bg = bg, main = "n_sd=3", las = 1, xlim = c(0.5, 2.5))
bg <- c(3, 2)[1 + CheckForOutliers(x, group, n_sd = 4, method = "logical")]
graphics::plot(x = as.numeric(group), y = x, pch = 21, cex = 3,
bg = bg, main = "n_sd=4", las = 1, xlim = c(0.5, 2.5))
graphics::par(mfrow = c(1, 1))

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

# no missing data in this matrix
all(is.finite(raw))

# check for outliers (computing n-fold sd distance from group mean)
tmp <- apply(raw, 2, CheckForOutliers, group = sam$GT, method = "dist") # plot a histogram of the observed distances graphics::hist(tmp, breaks = seq(0, ceiling(max(tmp))), main = "n*SD from mean", xlab = "n") # Calculate the amount of values exceeding five-sigma and compare with a standard gaussian table(tmp > 5) round(100 * sum(tmp > 5) / length(tmp), 2) gauss <- CheckForOutliers(x = rnorm(prod(dim(raw))), method = "dist") sapply(1:5, function(i) { data.frame("obs" = sum(tmp > i), "gauss" = sum(gauss > i)) }) # compare a PCA w/wo outliers RestrictedPCA( dat = raw, sam = sam, use.sam = sam$GT %in% c("Mo17", "B73"), group.col = "GT",
fmod = "GT+Batch+Order", P = 1, sign.col = "GT", legend.x = NULL, text.col = "Batch", medsd = TRUE
)
raw_filt <- raw
raw_filt[tmp > 3] <- NA
RestrictedPCA(
dat = raw_filt, sam = sam, use.sam = sam\$GT %in% c("Mo17", "B73"), group.col = "GT",
fmod = "GT+Batch+Order", P = 1, sign.col = "GT", legend.x = NULL, text.col = "Batch", medsd = TRUE
)



MetabolomicsBasics documentation built on May 29, 2024, 9:02 a.m.