splitPeaksFilter {SPUTNIK}

Test for the presence of split peaks.

Description

splitPeaksFilter returns a list of estimated split peak indices. Each element of the list contains an array of the original peak indices that can be merged. The name of the list element is the new m/z value associated with the merged peaks.

Usage

splitPeaksFilter(
  msiData,
  mzTolerance = 5,
  sharedPixelsRatio = 0,
  sparseness = "scatter.ratio",
  threshold = 0.5,
  returnDetails = TRUE,
  verbose = TRUE
)

Arguments

msiData

msi.dataset-class object. See msiDataset.

mzTolerance

numeric (default = 5). Maximum distance in PPM between the m/z values of two peaks to consider them for merging. See 'Details' section.

sharedPixelsRatio

numeric (default = 0). Maximum fraction of common pixels where the signal of two peaks is different from zero to consider them for merging. See 'Details' section.

sparseness

string (default = "scatter.ratio"). Method used to estimate the 'scatteredness' of the peak image. See 'Details' section.

threshold

numeric (default = 0.5). Threshold for scatteredness measure to consider peaks for merging. At least one of the merging peaks should have a measure associated with presence of structure.

returnDetails

logical (default = TRUE). Add details on merged peaks in the results.

verbose

logical (default = TRUE). Additional output text.

Details

splitPeaksFilter determines whether close peaks represent the same signal. This estimation is based on multiple conditions:

1. peaks m/z values should be closer than mzTolerance PPM

2. at least one of the peak images should be structured, accordingly to the sparseness measure. The threshold determines whether the pixel images are structured or not. The possible measures are:

   • "scatter.ratio": ratio between the number of non-zero pixels and the image size after binarization using Otsu's thresholding. A value close to 0 is associated with a more structured image, whereas a value close to 1 is associated with a less structured image. A suggested parameter of threshold = 0.5 represents the maximum value for this measure for a structured image. Minimum possible value is 1 / ( # non-zero pixels ).

   • "spatial.chaos": similar to the scatter ratio taking into account of the color histogram. A value close to 1 represents a structured image, whereas a value close to 0 represents a more scattered image. A suggested parameter of threshold = 0.8 represents the minimum value for this measure for a structured image. Maximum possible value is 1 - 1 / ( # histogram bins ). Here, we use the default number of bins equal to 30.

   • "gini.index": Gini index measures the image sparsity. A value close to 1 is associated with a sparse image whereas a value close to 0 is associated with a more uniform image. A suggested value of threshold = 0.9 represents the maximum value of this measure for a structured image.

3. the merged peaks image should be more structured than the single peak images, accordingly to the selected sparseness.

Value

peak.filter object. See applyPeaksFilter.

Author(s)

Paolo Inglese [e-mail]

References

Palmer, A., Phapale, P., Chernyavsky, I., Lavigne, R., Fay, D., Tarasov, A., ... & Becker, M. (2017). FDR-controlled metabolite annotation for high-resolution imaging mass spectrometry. Nature methods, 14(1), 57.

Hurley, N., & Rickard, S. (2009). Comparing measures of sparsity. IEEE Transactions on Information Theory, 55(10), 4723-4741.

Examples

## Load package
library("SPUTNIK")

## Mass spectrometry intensity matrix
X <- matrix(rnorm(200), 20, 40)
X[X < 0] <- 0

## Print original dimensions
print(dim(X))

## m/z vector
mzVector <- seq(600, 601, by = (601 - 600) / 39)

## Read the image size
imSize <- c(5, 4)

## Construct the ms.dataset object
msiX <- msiDataset(X, mzVector, imSize[1], imSize[2])

## Determine split peaks
sp.filter <- splitPeaksFilter(
  msiData = msiX, mzTolerance = 50,
  sharedPixelsRatio = 0,
  sparseness = "spatial.chaos", threshold = 0.5
)