1
XCMS / Optimizing centWave settings for HPLC-ESI-MS Orbitrap data
First post in this forum. I have used xcms in the past, but never before for Orbitrap data. I am having a heck of a time identifying centWave settings that yield acceptable-looking peaks, so I am hoping for some advice. I’ve reviewed a lot of the existing posts, but nothing has really done the trick. I have experimented around with ranges of different values for almost all of the centWave parameters, but still getting some rather questionable peaks. If anyone out there has any suggestions, I would be very grateful.
Background information: I am working with an 18-sample set of natural lipid extracts from an oxidation experiment. All samples run on an Agilent 1100 HPLC stack coupled to an Orbitrap Exactive Plus (100-1500 m/z scan range, chromatogram generally runs out to ~ 25 mins). I used an implementation of msconvert to convert and centroid the Thermo .raw data files, and extract + and – mode scans into separate files. I am looking now at the + mode data. I confirmed successful centroiding by inspection of the mzXML in several of the files with a text reader. I see:
Quote
<dataProcessing centroided="1">
</dataProcessing>
Right now, I am just working with a single xcmsRaw file, then generating plots using plotPeaks, in order to tune my settings as best as possible. I’ve already run an entire xcmsSet dataset containing all files through a full xcms workup, but I’m skeptical of the results since my peaks look so questionable.
Here’s what I have so far, and some example plots using plotPeaks() from this single sample. I started with the recommended centWave settings for HPLC/Orbitrap in Table 1 of Patti et al., 2012, "Meta-analysis of untargeted metabolomic data from multiple profiling experiment," Nature Protocols 7: 508-516. I’ve also experiment with massifquant, but massifquant appears to be identifying things that definitely aren’t peaks (confirmed by visual inspection when I use sleep = 0.01).
Code: [Select]
mzXMLfiles = list.files(mzXMLfiles_folder_pos, recursive = TRUE, full.names = TRUE)
# create xcmsRaw object from just a single sample (for method development, just using the first sample)
xfile_raw = xcmsRaw(mzXMLfiles[1])
profStep(xfile_raw) = 0.05
xfile_raw is:
Quote
An "xcmsRaw" object with 601 mass spectra
Time range: 0.1-1740.7 seconds (0-29 minutes)
Mass range: 100.0007-1499.9558 m/z
Intensity range: 849.698-406951000
MSn data on 0 mass(es)
with 0 MSn spectra
Profile method: bin
Profile step: 0.05 m/z (28000 grid points from 100 to 1499.95 m/z)
Memory usage: 149 MB
I then run:
Code: [Select]
rawpeaks = findPeaks.centWave(xfile_raw,
ppm = 2.5, # setting of Patti et al., 2012
peakwidth =c(10,45), # max lowered from Patti et al., 2012, HPLC recommended setting of 60 s based on visual inspection of sample data in Excalibur
fitgauss = TRUE,
noise = 5000,
# sleep = 1,
verbose.columns = TRUE,
snthresh = 10,
integrate = 1,
prefilter = c(3,10000) # 3.5k recommended by Patti et al. appears to be too low
# nSlaves = 4 # commenting out right now since this is just one sample
)
I receive the following output:
Quote
Detecting mass traces at 2.5 ppm ...
% finished: 0 10 20 30 40 50 60 70 80 90 100
43503 m/z ROI's.
Detecting chromatographic peaks ...
% finished: 0 10 20 30 40 50 60 70 80 90 100
15107 Peaks.
Warning message:
In .local(object, ...) :
It looks like this file is in profile mode. centWave can process only centroid mode data !
As I said, I have repeatedly confirmed that the data are centroided.
Here's where I am very skeptical of the peak-picking. Below are images of three sets of peaks from this list of rawpeaks; I am just choosing these three sets of 25 peak at random. I used this code:
Code: [Select]
plotPeaks(xfile_raw,rawpeaks[1:24,],figs = c(5,5),width = 100)
plotPeaks(xfile_raw,rawpeaks[2150:2174,],figs = c(5,5),width = 100)
plotPeaks(xfile_raw,rawpeaks[10150:10174,],figs = c(5,5),width = 100)
to generate these plots:
[attachment=2:cupyrl2t]Rplot.pdf[/attachment:cupyrl2t]
[attachment=1:cupyrl2t]Rplot2.pdf[/attachment:cupyrl2t]
[attachment=2:cupyrl2t]Rplot.pdf[/attachment:cupyrl2t]
For the first set, I am not reading too much into things; we (like everyone) have a lot of junk and noise that co-elutes early at low m/z. However, in some of the subplots, the peaks supposedly bounded by the grey lines aren't even visible. And, my settings seem to be picking many things that I wouldn't consider peaks.
I've worked up this same dataset using analogous settings in MAVEN (GUI application from the Rabinowitz lab at Princeton) and the vast majority of the peaks I see using that program are much better. So, I'm fairly certain it isn't something wrong with the underlying data (or our chromatography or MS settings). I am interested in using xcms, however, because we have a follow-on custom pipeline that's already scripted in R.
I appreciate any suggestions, and will very happily clarify anything I've written here if it's not clear.
Thank you in advance!
Jamie
[attachment deleted by admin]