Best program to calculate peak areas for GC/MS chromatogram

[seano]

For my research I obtain GC/MS chromatograms which involve many peaks (~30-40) and I need to find the area under these peaks. The file type for my chromatograms is .raw and I have been using Xcalibur to interpret my data. While Xcalibur can calculate area under peaks for me, the issue I am having is that it cannot save how I defined the area under my peaks. I would like to be able to manually integrate my peaks and then save how I integrated my peaks so I could make any edits at a later time. Does anyone know of a free software which can save peak integration? I could not figure out how to integrate using AMDIS, and I am now trying to see if this can be done using PARADISe. Any advice on this is appreciated! Thank you!

[lmh]

honestly, don't do it! Manual integration is a lot of work, and in any case, it's open to accusation that you've been biased in how you chose your baselines. And you're right, record-keeping is a pain.
The right way to solve this is to use the automated integration that Xcalibur provides. You've got a choice of multiple integrators, and it's almost always possible to find decent integrator settings that will integrate the peaks correctly. If you can't, it probably means that the peaks have such poor shape, such large noise, or such poor resolution, that they're not really suitable for quantitative work. But if you can integrate automatically, not only is it much quicker (whole batches of samples done in one go, and all the results in Excel at the click of a mouse!), but also, the processing method and its associated integrator settings are your record for the future. You can always recreate the exact same results from the same data file. Xcalibur's not the most sophisticated package out there, and I'm sure you'll find others that will do what you want, but if you turn to them, you're solving the wrong problem! (in my view... but it's only a personal view, no matter how strongly-held!)

[sbashkyrtsev]

Manual integration is ... open to accusation that you've been biased in how you chose your baselines.

I've heard this statement on multiple occasions, but.. When you modify parameters of an algorithm - isn't it just a manual integration with extra steps? In both cases you have some picture of a peak in your head and you're trying to make software "see" it in the same way. Sure, you have less flexibility, but still - what's really biased is the picture in your head, not the way you made software "see" it.

Another point that we could make: we strive for consistency. Even if the automatic algorithm and its configuration is biased - the bias is going to be the same across all chromatograms. But these algorithms are actually pretty bad at making consistent decisions - slight variations in the signal can yield a different picture. Humans seem to be much better in terms of consistency (well, at least when it's the same person).

Algorithms are much better at detecting "hidden" features like shoulders, round peaks. But once the software highlighted such feature - humans will still make final decisions based on their gut feeling and modify it either manually or through parameters.

[lmh]

I'm afraid I disagree. As you said, the difference between manual integration and automatic is that in manual you are doing individual things to each individual chromatogram, whereas in automatic you're doing the same to everything.
As to consistency, I'm not convinced that peak-finding algorithms have to be inconsistent. I think they can be inconsistent, particularly the older ones with their huge numbers of parameters, if they've been set up on a single file so that they find the right peaks for the wrong reasons. For example, where a chromatographer has just twiddled with parameters at random until it looks right, a peak might end at just the right place because there happened to be a little noise dip there. The integrator settings then fail on the next file, where there isn't a bit of noise in the right place, and the peak is far too wide. So the chromatographer reduces the peak width enormously and then on the third file, where there's a bit of noise half way through the peak, the integrator splits the peak because it now expects peaks that are only a quarter of the width that they really are, so it's desperate to find any excuse to split the peak... and so it goes on. But if the chromatographer had read the manual and realized that their integrator, in this case, was recognizing the end of the peak by its slope dropping below a certain threshold, then they'd have realized that back at the first file, they should have set up the slope properly, and it would have handled all three files. Really, it'd be great if manufacturers advertised their theory handbooks more widely, and if chromatographers read them!
But I do find modern integrators with far fewer parameters tend to be better, and more consistent. The algorithm designers have improved their stuff!

An edit: personally I rarely change my integrator parameters day-to-day or even month-to-month once I've got them set up. If the integrator found the peaks last year, it should still be finding them this year.

[seano]

Thank you both for your thoughts and the conversation. I noticed in Xcalibur there is 3 different peak detection algorithms: Genesis, ICIS, and Avalon. It seems there is also the option to edit the parameters for the algorithms a bit. I'll look into this as the set parameters for these algorithms don't give me great looking peaks. A lot of my peaks are reaching co-elution and have a fair amount of tailing. I think this may be why my adviser suggested manual integration. I'll make sure to post an update once I have found and settled on a method that works my purpose!

[sbashkyrtsev]

lmh, my main point was: if you don't like the results of automated peak detection, you change the parameters of the integrator. Which means you let your opinion dictate the shape of the peak and thus it's not too far from manual integration.

I understand that setting up parameters once and applying them to many injections is beneficial. Though if there are lots of injections (let's say 300) and 3 of them have some annoying artifact in the signal which makes them differ - it will be hard (or impossible) to come up with single configuration for the whole batch.. You'll have to change the parameters for those 3 chromatograms.

I've seen chemists apply a single integration method for batches and then going over all chromatograms and correcting manually those that they didn't like. Would you say that manual integration is optimal in such scenario or would you recommend updating integration method for each of them individually (which is more time consuming)?

As to consistency, I'm not convinced that peak-finding algorithms have to be inconsistent. I think they can be inconsistent, particularly the older ones with their huge numbers of parameters, if they've been set up on a single file so that they find the right peaks for the wrong reasons.

I'm really familiar only with ApexTrack, so can't judge the rest. But there seem to be only 2 underlying choices - either they use derivatives or fit models like Gaussian. I think almost all implementations of the 1st type (which seem to prevail) will fail in such situations:
* 2 very different peak profiles: one with tailing and one with fronting.
* Poor resolution of peaks and a lot of noise. Which makes you smooth more but after smoothing it won't see shoulders.
* A lot of noise and sharp peaks. Again requires a lot of smoothing which means boundaries are overestimated.

Though you did mention low-quality chromatograms in your 1st response. As far as I understood - you'd throw away these results and would try to improve separation method and re-run? Would you do the same if it were a qualitative or a semi-quantitative analysis?

personally I rarely change my integrator parameters day-to-day or even month-to-month once I've got them set up. If the integrator found the peaks last year, it should still be finding them this year.

Is it because you're striving for good separation method all the time which results in high-quality chromatograms? Or the software that you're using (which is it?) manages to recognize peaks well even if signal profile is very different?

Don't know if this is relevant to your situation - but I know some software packages that present user with the processed (smoothed) signal instead of the original one. I thinks this fools chemists into thinking that the integrator's choice of peak boundaries is correct, but if they saw the original signal - they'd probably disagree with some decisions of the integrator.

[lmh]

... I'm not going to bang on indefinitely because I'm just one opinion, and it'd be better to get a consensus view than over-weight my opinion.

But Seano, yes, if you've got coelution and really bad tailing, then you do have a problem. I personally believe that manual integration won't get you out of that problem, because with coelution different humans will draw the baseline in different places between two overlapping peaks (this is an exercise I do with an MSc class each year, except not this year!). Disentangling two overlapping peaks is actually quite tricky and I'm far from convinced that either the current state of automated software, or the human eye, does it very well. This goes beyond the reasonable bounds of this thread, but to my mind, the best way to handle overlapping, tailing peaks, would be to build a model for the peak shape of the two components by running separate standards, and model the actual sample chromatogram as a sum of the two known individual peak-shapes. A least squares best fit would then give the proportions. I do not know of any commercial chromatography software that does this. Probably it's because almost no chromatographer would trust it. The general view of respectable traditional chromatographers (a group to whose membership I hope one day to aspire) is that you should go away and improve your chromatography so that you get better resolution! But I know that can be extremely difficult.

The different algorithms in Xcalibur: Avalon was the original, and was designed for UV/Vis data. It can handle negative peaks, which Genesis and Icis don't like. ICIS was next, I think, and Genesis more recent. There is now another one too in Xcalibur in the latest Qualbrowser incarnation, FreeStyle, but weirdly it's not in the processing method bit??!! When in Xcalibur world I tend to gravitate towards Genesis but for no particular reason than that I know its parameters better than I do ICIS. Definitely find out how they work, they'll help you. But I don't think Xcalibur has the best algorithms. They've been around quite a while. Interestingly, Genesis and ICIS report S/N ratios orders of magnitude different for the same peak, so don't mix and match if you find yourself worried about S/N measurements, but that's another story.

Now Apex Track is an interesting one, because it uses 2nd derivatives, not just the slope, in an attempt to be more robust at finding peak starts and ends. Using the second-derivative, the curve, the rate-of-change-of-slope makes it fairly good at spotting shoulders. But there's a really silly thing I have to mention: ApexTrack is rather vulnerable to messing up because of how it's embedded in Waters' quantitative processing. You will know this as an experienced user, but just in case anyone ventures into this thread who doesn't know: if you open a whole chromatogram in MassLynx to have a look at it, ApexTrack starts at the beginning. If you use it in quantitative processing, it starts a short distance to the left of the peak, and stops a short distance to the right, to speed the process. But if it starts half way through another peak, it gets really confused, so it won't integrate the same as it did when you set it up... and that causes some thoroughly puzzled chromatographers. Fortunately you can change the range over which it integrates, it's an extra parameter in the integrator tab of TargetLynx, and presumably of QuanLynx (which I don't know).
Hmm, some interesting questions about habits with data-processing, and worthy of an answer: My attitude is that I don't update automated integration methods to deal with a handful of runs that I don't like. If most of the chromatograms are good, and a handful are poor, these are probably from samples that contained very little analyte (noise has dominated), samples that contained far too much of something (leading to chromatography artifacts) or otherwise samples with a problem. I personally regard the automated integrator's inability to handle the sample properly as an indication that the result is likely to be dubious whoever or whatever integrates it; ill-defined shoulders are never going to give you a precise result. I'm lucky in that in my line of business, usually people are only interested in a small result being small (they're going to plot it on some sort of graph, which will be dominated by the large results; if a small result is one pixel higher or lower on the page, no one is going to notice). So I don't much care if my integrator doesn't do a great job of a peak that's barely more than baseline noise. So I flag the bad runs, but I don't suppress or modify their results. I should perhaps rephrase my earlier message: in my view, the current generation of automated integrators does a good job on all peaks that are worth integrating, and where it fails, I don't personally trust myself (or any other human) to do much better. But I do appreciate not everyone has that luxury, and I'd love to see the world of chromatography be a bit more adventurous in how it untangles partially-overlapping peaks.
(sorry, that was a bit of an essay)

[James_Ball]

For our work doing environmental sample we understand that no one setting will integrate every chromatogram perfectly as one may be clean with a few peaks and others may have hundreds or thousands of peaks that interfere. We have a written protocol describing how a peak should be integrated when done manually and we are required to preserve a before and after image of the peak that is manually integrated so that in the future if there is any question there is the evidence of what was done to the chromatogram. (we need this since any of our work may some day go to court). As long as all the written procedures are followed and documented correctly you can keep questions to a minimum on your data. Most regulating agencies now days have this requirement and are satisfied if it is followed correctly.