Compounding (grouping) of LC-MS features
9 August 2024
Package
xcms 4.3.3
Package: xcms
Authors: Johannes Rainer
Modified: 2024-05-29 20:09:17.691102
Compiled: Fri Aug 9 20:12:24 2024
1 Introduction
In a typical LC-MS-based metabolomics experiment compounds eluting from the chromatography are first ionized before being measured by mass spectrometry (MS). During the ionization different (multiple) ions can be generated from the same compound which all will be measured by MS. In general, the resulting data is then pre-processed to identify chromatographic peaks in the data and to group these across samples in the correspondence analysis. The result are distinct LC-MS features, characterized by their specific m/z and retention time range. Different ions generated during ionization will be detected as different features. Compounding aims now at grouping such features presumably representing signal from the same originating compound to reduce data set complexity (and to aid in subsequent annotation steps). General MS feature grouping functionality if defined by the MsFeatures package with additional functionality being implemented in the xcms package to enable the compounding of LC-MS data.
This document provides a simple compounding workflow using xcms. Note that the present functionality does not (yet) aggregate or combine the actual features per values, but does only define the feature groups (one per compound).
2 Compounding of LC-MS data
We demonstrate the compounding (feature grouping) functionality on the simple toy data set used also in the xcms package and provided through the faahKO package. This data set consists of samples from 4 mice with knock-out of the fatty acid amide hydrolase (FAAH) and 4 wild type mice. Pre-processing of this data set is described in detail in the main vignette of the xcms package. Below we load all required packages and the result from this pre-processing updating also the location of the respective raw data files on the current machine.
library(MSnbase)
library(xcms)
library(faahKO)
library(MsFeatures)
xmse <- loadXcmsData("xmse")Before performing the feature grouping we inspect the result object. With
featureDefinitions we can extract the results from the correspondence
analysis.
featureDefinitions(xmse) |> head()## mzmed mzmin mzmax rtmed rtmin rtmax npeaks KO WT peakidx
## FT001 200.1 200.1 200.1 2902.634 2882.603 2922.664 2 2 0 458, 116....
## FT002 205.0 205.0 205.0 2789.901 2782.955 2796.531 8 4 4 44, 443,....
## FT003 206.0 206.0 206.0 2789.405 2781.389 2794.219 7 3 4 29, 430,....
## FT004 207.1 207.1 207.1 2718.560 2714.047 2727.347 7 4 3 16, 420,....
## FT005 233.0 233.0 233.1 3023.579 3015.145 3043.959 7 3 4 69, 959,....
## FT006 241.1 241.1 241.2 3683.299 3661.586 3695.886 8 3 4 276, 284....
## ms_level
## FT001 1
## FT002 1
## FT003 1
## FT004 1
## FT005 1
## FT006 1Each row in this data frame represents the definition of one feature, with its
average and range of m/z and retention time. Column "peakidx" provides the
index of each chromatographic peak which is assigned to the feature in the
chromPeaks matrix of the result object. The featureValues function allows to
extract feature values, i.e. a matrix with feature abundances, one row per
feature and columns representing the samples of the present data set.
Below we extract the feature values with and without filled-in peak data. Without the gap-filled data only abundances from detected chromatographic peaks are reported. In the gap-filled data, for samples in which no chromatographic peak for a feature was detected, all signal from the m/z - retention time range defined based on the detected chromatographic peaks was integrated.
head(featureValues(xmse, filled = FALSE))## ko15.CDF ko16.CDF ko21.CDF ko22.CDF wt15.CDF wt16.CDF wt21.CDF
## FT001 NA 506848.9 NA 169955.6 NA NA NA
## FT002 1924712.0 1757151.0 1383416.7 1180288.2 2129885.1 1634342.0 1623589.2
## FT003 213659.3 289500.7 NA 178285.7 253825.6 241844.4 240606.0
## FT004 349011.5 451863.7 343897.8 208002.8 364609.8 360908.9 NA
## FT005 286221.4 NA 164009.0 149097.6 255697.7 311296.8 366441.5
## FT006 1160580.5 NA 380970.3 588986.4 1286883.0 1739516.6 639755.3
## wt22.CDF
## FT001 NA
## FT002 1354004.9
## FT003 185399.5
## FT004 221937.5
## FT005 271128.0
## FT006 508546.4head(featureValues(xmse, filled = TRUE))## ko15.CDF ko16.CDF ko21.CDF ko22.CDF wt15.CDF wt16.CDF wt21.CDF
## FT001 135162.4 506848.9 111657.3 169955.6 209929.4 141607.9 226853.7
## FT002 1924712.0 1757151.0 1383416.7 1180288.2 2129885.1 1634342.0 1623589.2
## FT003 213659.3 289500.7 164380.7 178285.7 253825.6 241844.4 240606.0
## FT004 349011.5 451863.7 343897.8 208002.8 364609.8 360908.9 226234.4
## FT005 286221.4 285857.6 164009.0 149097.6 255697.7 311296.8 366441.5
## FT006 1160580.5 1102832.6 380970.3 588986.4 1286883.0 1739516.6 639755.3
## wt22.CDF
## FT001 138341.2
## FT002 1354004.9
## FT003 185399.5
## FT004 221937.5
## FT005 271128.0
## FT006 508546.4In total 351 features have been defined in the present data set, many of which most likely represent signal from different ions (adducts or isotopes) of the same compound. The aim of the grouping functions of are now to define which features most likely come from the same original compound. The feature grouping functions base on the following assumptions/properties of LC-MS data:
- Features (ions) of the same compound should have similar retention time.
- The abundance of features (ions) of the same compound should have a similar pattern across samples, i.e. if a compound is highly concentrated in one sample and low in another, all ions from it should follow the same pattern.
- The peak shape of extracted ion chromatograms (EIC) of features of the same compound should be similar as it should follow the elution pattern of the original compound from the LC.
The main method to perform the feature grouping is called groupFeatures which
takes an xcms result object (i.e., an XcmsExperiment or XCMSnExp) as
input as well as a parameter object to chose the grouping algorithm and specify
its settings. xcms provides and supports the following grouping approaches:
SimilarRtimeParam: perform an initial grouping based on similar retention time.AbundanceSimilarityParam: perform a feature grouping based on correlation of feature abundances (values) across samples.EicSimilarityParam: perform a feature grouping based on correlation of EICs.
Calling groupFeatures on an xcms result object will perform a feature
grouping assigning each feature in the data set to a feature group. These
feature groups are stored as an additional column called "feature_group" in
the featureDefinition data frame of the result object and can be accessed with
the featureGroups function. Any subsequent groupFeature call will
sub-group (refine) the identified feature groups further. It is thus possible
to use a single grouping approach, or to combine multiple of them to generate
the desired feature grouping in an incremental fashion. While the individual
feature grouping algorithms can be called in any order, it is advisable to use
the EicSimilarityParam as last refinement step, because it is computationally
very expensive, especially if applied to a result object without pre-defined
feature groups or if the feature groups are very large.
2.1 Grouping of features by similar retention time
The most intuitive and simple way to group features is based on their retention time. Before we perform this initial grouping we evaluate retention times and m/z of all features in the present data set.
plot(featureDefinitions(xmse)$rtmed, featureDefinitions(xmse)$mzmed,
xlab = "retention time", ylab = "m/z", main = "features",
col = "#00000080", pch = 21, bg = "#00000040")
grid()
Figure 1: Plot of retention times and m/z for all features in the data set
Several features with about the same retention time (but different m/z) can be spotted, especially at the beginning of the LC. We thus below group features within a retention time window of 10 seconds into feature groups.
xmse <- groupFeatures(xmse, param = SimilarRtimeParam(10))The results from the feature grouping can be accessed with the featureGroups
function. Below we determine the size of each of these feature groups (i.e. how
many features are grouped together).
table(featureGroups(xmse))##
## FG.001 FG.002 FG.003 FG.004 FG.005 FG.006 FG.007 FG.008 FG.009 FG.010 FG.011
## 4 6 2 4 6 5 2 4 3 4 3
## FG.012 FG.013 FG.014 FG.015 FG.016 FG.017 FG.018 FG.019 FG.020 FG.021 FG.022
## 6 4 2 3 3 3 4 3 2 5 2
## FG.023 FG.024 FG.025 FG.026 FG.027 FG.028 FG.029 FG.030 FG.031 FG.032 FG.033
## 2 4 5 2 5 2 3 4 2 2 4
## FG.034 FG.035 FG.036 FG.037 FG.038 FG.039 FG.040 FG.041 FG.042 FG.043 FG.044
## 2 2 6 3 2 2 4 2 2 4 4
## FG.045 FG.046 FG.047 FG.048 FG.049 FG.050 FG.051 FG.052 FG.053 FG.054 FG.055
## 2 3 2 3 3 5 2 3 4 2 6
## FG.056 FG.057 FG.058 FG.059 FG.060 FG.061 FG.062 FG.063 FG.064 FG.065 FG.066
## 2 6 4 2 4 3 3 2 2 2 2
## FG.067 FG.068 FG.069 FG.070 FG.071 FG.072 FG.073 FG.074 FG.075 FG.076 FG.077
## 3 2 2 3 2 3 2 3 3 3 3
## FG.078 FG.079 FG.080 FG.081 FG.082 FG.083 FG.084 FG.085 FG.086 FG.087 FG.088
## 2 3 3 3 2 3 3 3 3 3 2
## FG.089 FG.090 FG.091 FG.092 FG.093 FG.094 FG.095 FG.096 FG.097 FG.098 FG.099
## 3 2 2 2 2 3 2 4 2 3 3
## FG.100 FG.101 FG.102 FG.103 FG.104 FG.105 FG.106 FG.107 FG.108 FG.109 FG.110
## 3 3 2 2 2 2 2 2 2 2 2
## FG.111 FG.112 FG.113 FG.114 FG.115 FG.116 FG.117 FG.118 FG.119 FG.120 FG.121
## 2 2 1 1 1 1 1 1 1 1 1
## FG.122 FG.123 FG.124 FG.125 FG.126 FG.127 FG.128 FG.129 FG.130 FG.131 FG.132
## 1 1 1 1 1 1 1 1 1 1 1
## FG.133
## 1In addition we visualize these feature groups with the plotFeatureGroups
function which shows all features in the m/z - retention time space with grouped
features being connected with a line.
plotFeatureGroups(xmse, pch = 21, lwd = 2, col = "#00000040",
bg = "#00000020")
grid()