When using xcms to group peaks with PeakDensityParam, the resulting feature RT is currently calculated as the median of the grouped peaks. In datasets where:
- Low-intensity peaks are present in most samples, and
- High-intensity peaks are present in fewer samples,
the median RT can be pulled away from the major (intense) peaks, resulting in inaccurate feature RT assignments.
Example:
- 30 samples
- 20 small peaks around 910 s (low intensity, present in most samples)
- Major peaks around 870 s (high intensity, 1000× stronger, present in 10 samples)
Current behavior: grouped feature RT → ~903 s (median of all peaks), which does not reflect the major peak RT.
Current workaround:
For m/z in CentWaveParam, there is mzCenterFun = "wMean" to calculate an intensity-weighted m/z, but no equivalent parameter exists for MZ or RT in PeakDensityParam.
Suggestion:
Add rtCenterFun and optionally mzCenterFun parameters to PeakDensityParam / groupChromPeaks with options like:
"median" (current default)"mean""wMean" (weighted by intensity)
This would allow feature RT (and optionally m/z) to better reflect major/intense peaks, improving grouping accuracy, especially in heterogeneous datasets.
Benefits:
- Prevents feature RT from being skewed by numerous low-intensity peaks
- Makes grouped features more representative of biologically relevant peaks
- Consistent interface with
CentWaveParam’s mzCenterFun
Note on relationship to bw:
The proposed rtCenterFun parameter is independent of bw. While bw controls the density bandwidth and thus whether peaks are merged or split
- Increasing
bw → broader grouping → risk merging distinct peaks
- Decreasing
bw → narrower grouping → risk splitting the same peak across groups
rtCenterFun would only determine how the RT of an already grouped feature is calculated (e.g., median, mean, intensity-weighted mean). This allows users to maintain their desired grouping behavior without being forced to adjust bw to compensate for RT skew caused by uneven peak intensities.
When using
xcmsto group peaks withPeakDensityParam, the resulting feature RT is currently calculated as the median of the grouped peaks. In datasets where:the median RT can be pulled away from the major (intense) peaks, resulting in inaccurate feature RT assignments.
Example:
Current behavior: grouped feature RT → ~903 s (median of all peaks), which does not reflect the major peak RT.
Current workaround:
For m/z in
CentWaveParam, there ismzCenterFun = "wMean"to calculate an intensity-weighted m/z, but no equivalent parameter exists for MZ or RT inPeakDensityParam.Suggestion:
Add
rtCenterFunand optionallymzCenterFunparameters toPeakDensityParam/groupChromPeakswith options like:"median"(current default)"mean""wMean"(weighted by intensity)This would allow feature RT (and optionally m/z) to better reflect major/intense peaks, improving grouping accuracy, especially in heterogeneous datasets.
Benefits:
CentWaveParam’smzCenterFunNote on relationship to
bw:The proposed
rtCenterFunparameter is independent ofbw. Whilebwcontrols the density bandwidth and thus whether peaks are merged or splitbw→ broader grouping → risk merging distinct peaksbw→ narrower grouping → risk splitting the same peak across groupsrtCenterFunwould only determine how the RT of an already grouped feature is calculated (e.g., median, mean, intensity-weighted mean). This allows users to maintain their desired grouping behavior without being forced to adjustbwto compensate for RT skew caused by uneven peak intensities.