Acceptance criteria for peak to valley ratio

[Bridget Sykes]

Hi all!
I have the same question about peak to valley ratio acceptance criteria as previously posted.... what is an appropriate acceptance criteria for peak to valley ratio controlling a non-baseline, but still adequately resolved peak?

I saw a reply stating 1.5 was appropriate - any advice as to where this comes from (eg any regulatory ref)?

Any advice on this is much appreciated!

[mbicking]

If you search through the old threads, you can see this has been debated for some time. The original source may be hard to find though. But the idea is that the classic resolution equation is really not appropriate when the peaks are not the same size. If one peak is significantly bigger than the other, they can have a calculated resolution of 1.5 ("baseline resolution"), but the valley can be well above "baseline." The peak to valley ratio is another way to assess the effective separation; that is, how far does the valley go below the smaller peak. A larger ratio implies a better separation.

However, this is just another poor attempt to take a complex concept and put it into a simple yes/no format - something the regulators like. Ultimately, the way you draw the integration baselines will have a far greater impact on the accuracy. And the best options for drawing the baseline will change with the peak size ratio and resolution. So, ther isn't a simple answer to the question.

[lmh]

This isn't an answer, but is closely related:

A while back, I saw a PhD project on quantification of peaks; the author had looked at modelling hplc peaks as tailing Gauss curves, which can be described by an area, retention time, width, and tailing factor (i.e. 4 parameters). Given two overlapping peaks, it was possible to do a best fit to two such curves (8 parameters). Two of the fitted parameters (the two areas) are then exactly what you want to measure. Although fitting a curve with as many as 8 parameters ought to scare anyone, it's not quite as scary as an 8-parameter polynomial fit (which could probably be made to model a longitudinal section of the Alps); since we're obliged to have peaks within certain constrained sizes, the model isn't so likely to produce rubbish answers. Of course there's no need to fit the entire chromatogram at once; you'd just fit bits between baseline.

This struck me as a very powerful way to quantify mixed peaks, much more elegant than dropping lines from valleys, or skimming peaks on the side of others (which peak is riding on the side of which??), all of which are frankly botched approximations. And yet it's not there in the commercial software.

Is this just that I'm looking at the wrong packages? Was it actually a bad idea? Or is it just that the manufacturers haven't felt it worthwhile?

[Rob Burgess]

Two points:

Bridget - If you do want a quantifiable figure your better off using discrimination factor, as this "normalises" your peak to valley ratio. Using this value you get values between 0 and 1 for the smaller peak (where 1 is complete baseline separation).

lmh - is there any further info on the method your are talking about here?

[HW Mueller]

A long time ago I bought a curve fitting and resolution software and found out that nature was much more tricky and varient than the software. I never used it.
The way around the problem is usually to use peak hight instead of area. (But don´t forget that you have to do the standards, etc., with peak hight also.) Furthermore, nobody forbids the trial of both methods.

[lmh]

Rob Burgess, I'm mortified, I can't find the reference. I'm sure I wouldn't have thrown it away, but being a thesis I doubt I printed it (or much of it). Not a good reflection on my QA, I know. I'll keep looking.

However, I think maybe I was being harsh to software-writers. This seems old-hat in non-hplc places, e.g.:

http://people.bu.edu/tullius/paper_pdfs/GelExplorer.pdf

Doing a quick Google search I found a couple of hplc curve-fitting examples, but it sounds as though HW Mueller has greater experience. Personally I've never tried this, as I tend to set up methods using instruments software rather than third-party software.

[Shreekant]

Hi Merlin,

I mostly agree with you, but would like to add a couple of points. I think peak to valley ratio resolution should be used when the chromatographic software per the conventional resolution either a) fails to calculate the resolution due to its inability to calculate peak at half height which is a result of small peak fails to come down below the half height or b) the conventiona resolution value is less than, I think 1.5.

I look forward to hearing from you.