Skip to contents

Calculate a distance matrix between MS/MS spectra in the MetFamily feature Matrix

calculateDistanceMatrix.Rd

Computes a distance matrix between MS/MS spectra in the feature matrix representation using various distance or similarity measures. The function supports multiple distance metrics, of which the following are also available in the MetFamily GUI: "Jaccard", "Jaccard (intensity-weighted)", "Jaccard (fragment-count-weighted)" and "NDP (Normalized dot product)". The following list defines the metrics and their characteristics:

Usage

calculateDistanceMatrix(
  dataList,
  filter,
  distanceMeasure = "Jaccard",
  progress = FALSE
)

Arguments

dataList

List object containing precursor, feature and MS/MS data.

filter

Logical or integer vector indicating for which precursors to include the MS/MS spectra.

distanceMeasure

Character string specifying the distance metric to use. Supported values include "Jaccard", "Manhatten", "NDP (Normalized dot product)", and others.

progress

Logical or NA. If TRUE, progress is reported via incProgress().

Value

A list with elements:

distanceMatrix

A numeric matrix of pairwise distances.

filter

The filter vector used, same as the input parameter.

distanceMeasure

The distance metric used, same as the input parameter.

Details

  • "Jaccard": Jaccard distance \(1 - |A \cap B| / |A \cup B|\): the fraction of matching fragments among all fragments, where A and B are MS/MS features of two different precursors.

  • "Jaccard (intensity-weighted)": Jaccard distance weighted by the intensity of features. First, intensities are discretized: [0.01-0.2[ are set down to 0.01, [0.2, 0.4[ down to 0.2, and >= 0.4 increased to 1. For matching fragments the higher intensity is used. Then, Jacard becomes the sum of matching intensities among the sum of all intensities in A and B. \(1 - \sum_{i\in matches} max(A_i, B_i) / (sum(A) + sum(B))\)

  • "Jaccard (fragment-count-weighted)": Jaccard distance weighted by relative occurance of the fragments among the precursors after filtering counts: \(1 - \sum_{i\in matches} freq(A_i) / (sum_{\notin matches} freq(A) + sum_{\notin matches}(B))\)

  • "NDP (Normalized dot product)": Normalized dot product similarity: \(NDP = \frac{\left( \sum_{i}^{\text{S1\&S2}} W_{\text{S1},i} W_{\text{S2},i} \right)^2}{\sum_i W_{\text{S1},i}^2 \sum_i W_{\text{S2},i}^2}\) as described in Gaquerel et al. 2015. (10.1073/pnas.1610218113)https://www.pnas.org/doi/10.1073/pnas.1610218113#sec-4-5