Pro Scientific Opinions Sought: Integrate as one or two?

[Pelgy2]

Hello,

I just completed an oral interview for a new position after passing the written exam.
One of the questions during the interview involved a picture of a chromatogram like that from tinypic pasted below.


[/img][/url]

The question was, how would I integrate the peaks, based upon my personal opinion.
I gave my answer and found out on Friday that I failed the interview due to my answer to that question.

I said that the peaks were not resolved enough, so I would integrate across the bottom of both and report it as a coelution, Compound1-Compound2.
Unfortunately, I don't remember the retention times or area counts of the peaks so I can't do a resolution calculation, after the fact. I know it should be 1.5 or 2, minimum.

But I would be interested in some outside professional opinions on whether the two peaks are resolved enough to integrate them as they said they could be, as I've drawn in the tinypic. They would draw a line down the middle where the cleavage is and integrate the peaks separately.

Thanks for your time,
Peg

[bhuvfe]

I would have integrated those peaks as they said. It's a quite common way if
the peaks are of similar height.

You may find this article interesting; http://chromatographyonline.findanalyti ... 9&pageID=1

bhuvfe

[Pelgy2]

Hello,

Thanks very much for replying.

I checked out the link. Can you really believe someone with a hair-do like that?! (joke)
Thanks again for the info.

Any other opinions out there?

Peg

[Consumer Products Guy]

I'd also drop a line down and integrate separately, about the best one can do here.

And my hair is not much different than his...

[bisnettrj2]

Sacrilege! No one insults the 'do of the one true John Dolan! Oh, and especially if one has as thick and vibrant a head of hair as shown in that pic. (Reminds self - time to shave balding head again...)

In re: the issue at hand... I'd do the baseline drop like eveyone else, but with a caveat - if your company has an SOP that states a minimum resolution requirement in order to integrate individually a pair of peaks, and this particular set does not meet that requirement, then I think you would be within your rights to integrate this as one peak. However, I think the requirements for such a situation would have to be relatively specific, and in general, one would treat the example as two peaks.

[Peter Apps]

I am puzzled as to what the point would be of integrating both peaks together - clearly they are not the same compound so what does the combined area tell you about the sample ? If you only want to know the total content of the two compounds then you do not need to separate them at all, either chromatographically or with the integration.

Peter

[HW Mueller]

The only possibility I see here is that one does this via peak hight, but then one needs to calibrate via peak hight first.
I don´t trust peak fitting which would assume that the peaks are Gaussian.

[bisnettrj2]

I could see integrating them together if they are related, perhaps isomers (I have two pairs in the analysis I run that almost look like this pair), and if the resolution between isomers isn't sufficient (according to either a resolution equation that is pre-defined, or maybe the famed 'calibrated eyeball'), and one isomer is present at a significant quantity as compared to the other (making it difficult to quantitate both peaks), then the method calls for the analyst to report the result as the sum of both peaks (and as the sum of both isomers). However, the case shown is the reason I quant by peak height and not area, and would have to be narrowly defined in order to be integrated as one peak, as the OP did. I think, however, that most of us are saying - two peaks, not one.

[Uwe Neue]

Hans, it does not matter whether you trust the fitting algorithms or not, but they definitely will do a better job than quantitation by peak height.

[bisnettrj2]

Uwe/HW Mueller - what's 'peak-fitting'? I found a couple software packages online, and it basically looks like the next evolution of smooting? And why will it do a "better job than quantitation by height"?

[Uwe Neue]

All methods have their limitations, but if you fit a Guassian or even a non-Gaussian peak shape to a pair of overlapping peaks, you get a truer picture of the peak areas than from a simple determination by either a vertical line or a by peak height. You can try this yourself (for simple Gaussian peaks) in a spreadsheet...

[DR]

If on an interview, my answer would be "check w/ supervision" as clearly there is either a lack of resolution between 2 analytes or a column issue resulting in the splitting of a single analyte. Either way, supervision should have a hand in deciding if it should be integrated at all, or thrown out and rerun...

If forced to integrate, I'd either do a drop down or an exponential skim (the later being preferable if the resulting integration looked reasonable and could be done consistently - assuming I had several chromatograms looking like the one in question).






So, am I hired?

[HW Mueller]

Some 20 years before I saw the following statement by Dyson I put a peak fitting software on the shelf.


From

tom jupille

Joined: 11 Aug 2004
Posts: 2421
Location: Walnut Creek, CA / USA
Posted: Sat Dec 26, 2009 7:31 am Post subject:

--------------------------------------------------------------------------------

Unfortunately, there is no hard-and-fast number. "Baseline" resolution of 1.5 was originally defined that way as the resolution at which you have just less than 1% overlap between two equal-sized, Gaussian peaks. The FDA suggests that resolution should be greater than 2. Where there is a big disparity in peak sizes or serious tailing, even greater resolution may be required. Perhaps the best statement comes from Dyson's book on integration:

“. . . errors arising from peak overlap are introduced by the algorithms of perpendicular and tangent separation and cannot be eliminated by anything but better chromatography. Integrators are able to generate a highly precise and totally inaccurate set of results for all the foregoing examples.â€

[lmh]

(1) the ranty bit:

I have to strenuously disagree about peak fitting. Currently, peak fitting may not work, but I think it's not useful to abandon it on the assumption that it never will.

The major problem in peak-fitting (apart from computational demand) is the assumption of some particular peak shape, but seeing as we have a lot of measured points, we shouldn't have to assume the shape. We should be able to use the clues we have to work out what the shape genuinely is, albeit given imperfect data from mixed peaks.

If two things coelute, then from a theoretical perspective neither height nor area-with-a-drop-line will do, because both include area/height due to the wrong analyte. One might be better than the other, but neither is actually "right".

Of course no one will (or should) attempt to compute themselves out of bad chromatography unless they have to, but what are we to do when we either run out of cash (we cannot afford a 2 hour run that separates the peaks) or run out of options (we simply cannot separate the peaks with any known column or gradient)? If we genuinely can't separate the peaks, we have to do the best we can and report the dangers. It's hard to justify a drop-line as being any better (or worse) than peak-fitting.

Theoretically, again, both area-with-a-drop-line and height reduce our data to two points, so they don't have much information about peak shape to play with. The peak-fitting approach uses all the data, so it ought to be able to do a better job. Whether it does, at the moment, is another question, and a doubtful one.

(2) what I'd do:

If I were following a SOP, I'd do what it said: report the peaks with a drop line if that's requested, reject them if the resolution doesn't match what is demanded, whatever.

If I were developing a new method, I would report to the client that in my view resolution was inadequate, and offer them the options of:
(1) report isomers together as a combined area
(2) try to improve resolution by changing method
(3) use drop-lines, skims or peak-height, but always with the proviso that the results are unlikely to be so reliable as with good resolution.

In case (3) I would ideally like extra evidence (for example from samples spiked with known amounts of expected analyte) that the approach isn't giving unacceptable errors.

Practically, I'd also like to check that the resolution is genuinely bad, and not influenced by poor connections, big dead volumes, bad choice of injection solvent, or anything else that can easily be cleared up.

[PhotonicGuy]

I suppose if I would have been at an oral exam I would also have chosen to integrate them separately.