Hy all!
I have a question regarding fillPeaks(). Do the intensities coming from fillPeaks have a SN-value of NA?
I'm asking because when I look at my data I have a lot of NA's in my SN column and I want to avoid them since the method xcmsSet should identify most of my peaks. When I look at the EIC-diagrams there are very clear peaks but a lot of them come (as I think) from fillPeaks and not from xcmsSet-method.
I'm using the following parameters:
xset <- xcmsSet(method='centWave', ppm=15, peakwidth=c(30,120), snthresh=100, prefilter=c(3,10000), mzCenterFun="wMean", integrate=1, mzdiff=-0.001, fitgauss=TRUE, verbose.columns=TRUE, nSlave=4)
Second question would be: Which parameters should i change to get more Peaks from xcmsSet or to get more SN-values which are not NA?
br
Gunnar
PS: We are using a LTQ Orbitrap XL for analysis
[attachment deleted by admin]
Your are using two intensity related thresholds
1. S/N ratio = 100
2. prefilter = 3*10,000 = 30,000 minimum intensity
You can try to lower these thresholds to get more features from feature detection and not fillPeaks.
fillPeaks works independently of the feature detection method that was used before
by integrating intensities within the feature group boundaries (m/z and RT).
This method does not calculate comparable S/N values for these features that get filled up.
I lowered snthresh to 50 and prefilter to c(3,3333) and that helped.
Thanks a lot
Gunnar
I have not clear in my mind how prefilter and snthresh works together. So, which filter is master and which is slave?? I can I control their interaction?
"prefilter" and "noise" are filters that have an effect very early in the process, during the detection of ROI's.
A signal/noise level is not know at this point.
S/N is calculated as one the last steps when all feature characteristics - including the noise level - are known.
So it right to assume that prefilter works fisrt and than the S/N filter.
Using the example of gunnar:
First the prefilre select features accordining to
c(3, 3333), three peaks with an intensity of 3333 counts each, than the S/N filter exclude all the peaks below a s/n of 50.
Is that right??
Yes it will keep only ROI's that have more than 3 centroids with intensity > 3333. (I guess Gunnar choose these values because he wants to have ROI's with a total intensity of 10000 or more.)
Actually, I was thinking of replacing the three different prefilter options (prefilter (k,I) and noise) by just one parameter - the minimum total intensity of the ROI.
The S/N filter in this example will exclude features with S/N < 50 (which is pretty high).
Ralf wrote: I'm just getting really started on metabolomics work (well, restarted), and so I am unclear on how to proceed with these NAs. I wonder if it is appropriate to estimate intb and sn for these peaks whose intensities (into, maxo) are estimated by fillPeaks? What is your opinion and what do you do for your own work?
I'm not sure I understand why you need the S/N of features that were filled in by fillPeaks.
Since they were not detected at the first place, they are below the S/N threshold by definition.
fillPeaks does not really detect features, it merely integrates intensity values in the given m/z/RT range, which might just be pure noise.
This makes sense for features that are simply not present in one of the groups,e.g. a knock out.
If you are only interested in features that are above a certain S/N threshold (i.e. detected and not filled in)
you can group using minfrac=1 - then you shouldn't see any NA's.
Hi Ralf,
Thanks, that answers my question about S/N. What is your approach with respect to intb? Is there an argument to be made for not filling in intb or is that an outstanding error/issue?
fillPeaks does not fill in intb values, that is an outstanding issue with no solution in sight, yet.
We typically use into values for comparisons and maxo values for things like isotopic peak ratios.
I was leaning the same way. Awesome...