I have designed a quick macro to go through and re-process a subset of my 2D data.
Essentially, I wanted to think about it like you might for a 1D spectra, such that MI could be set by the macro (say to 0.001234), but since this is setting relative thresholds, I either get 12 peaks or 8900+, depending on the intensity of the solvent peaks. Since I'm working with mixtures, the dynamic range of what I'm looking into can vary.
Hi Joseph,
YMAX_P is the intensity of the tallest peak, S_DEV is the standard deviation of the spectrum. YAMX_P/S_DEV*n with n = 10 - 100. will give you ca good number to enter into MI.
Clemens
Clemens, as always, offers a good approach. The difficulty in automating peak picking (and integration too) is that the context determines what is the right value to set. As scientists, we look at a spectrum and make some assessment of it based on what outcome we require. This is really hard (impossible?) for an algorithm to reproduce because it does not have the context to "know what is best".
In a similar case, I have spent a lot of time trying to work out the best vertical scaling algorithm for creating high quality PDFs of 1D proton spectra (to go into lab notebooks, archives). My current algorithm does an excellent job 80% of the time, a good job 18% of the time and an unsatisfactory job 2% of the time (based on subjective review). I don't believe that it is possible to improve it much further for the reasons outlined above...
Thanks John, this like many other things in life will be subject to the 80.20 rule.
Joseph, I forgot to add that the parameter PPMPNUM can be used to set the maximum number of points. The list will be limited from the top down. You could set this to 8899 ;-)
Clemens, as always, thanks for the input and suggestions. Could you please clarify how to calculate and set this value as MI in the macro itself? I can retrieve YMAX_P and S_DEV values manually via the command line, no problem, but how can these values be combined into an equation the macro can understand, and what is meant by n in the equation? In a test spectra, the YMAX_P/S_DEV equated to roughly 8000, which would suggest that n is a scalar? Are the units of YMAX_P and S_DEV in points, and if so, does n convert the ratio to a relative value? Or have I misinterpreted this?
Just to clarify, I'm writing this through edmac, not through edau. Is this my problem?
I will also limit PPMPNUM as well to ensure a personal victory.
John, I have been struggling with these types of thought experiments, which is why I'm revisiting some of the spectra. Unsurprisingly, I am beginning to think that perfection is the enemy of completion, and would like to find a way to not have to peak pick hundreds of spectra by hand anymore.
Joseph, it is more than a thought experiment Our systems generate over 100,000 spectra per year. An 80:20 rule leaves us with 20,000 spectra to manually correct. My quest has been to steadily move this percentage upwards so people don't waste their time on this. I am now at 98% and, fairly obviously, further improvement is much harder (or impossible) to achieve.
As a matter of interest, what do you use the picked peaks for? We don't pick peaks by default but have a plot that has expansions allowing accurate measurement of chemical shift and splittings. For a more complete solution, we use either ACD Spectrus or MNova to perform a more complete data analysis (multiplet analysis, etc.). Even peak-picking algorithms will struggle with shoulders on peaks. The MNova Global Spectral Deconvolution (GSD) algorithm does a good job of going beyond simple maxima and extracting more complete information.
I was thinking of an AU program to set MI. If in your example The largest peak YMAX_p/S_DEV gives ca 8000. Then MI could be set to 1/8000*n or
0.000125*n. With n=10 MI would end up being 0.00125 with n=100 it is 0.0125. The optimum setting in order to get a large number of the peaks of interest and not too much will be somewhere in between.
you could write something like this:
int ymaxp;
float sdev, mi;
GETCURDATA
FETCHPARS ("YMAX_P", &ymaxp);
FETCHPARS("S_DEV", &sdev);
mi=sdev/ymaxp*100;
STOREPAR("MI",mi);
QUIT
- Badges
- Science topic
- Similar topics
- Mathematics