Thank you so much for your suggestions and comments on quantitative NMR.
Here is a summary of responses with names and personal notes removed.
ORIGINAL REQUEST:
Dear AMMRL,
Do any of you currently use software (or macros your lab has created) to
improve integration of your 1-D spectra? What is the name of the software
and who made it? How did you evaluate its performance? What is your
application for quantitative NMR?
I have been doing a lot of quantitative analysis by proton NMR (standards
authentication, etc.) and have written some macros to make it easier for me
to quantitate the commonly seen compounds. However, I am about to start on
some work that will require I do a lot better. I have wondered what the
limitation of NMR is for accuracy and precision.
Also, can any of you recommend a professor or chemist who is working on
improvement of NMR integration?
Any help you can give me is much appreciated.
Patrick Hays
*************
Response #1:
You might try LC-Model:
http://s-provencher.com/
The manual gives some insight to the issues concerning the types of
measurments that you are trying to do.
It is not meant to do High Resolution stuff but it can be coaxed to do so.
Other wise MRUI:
http://www.mrui.uab.es
Should get it.
Things to consider when thinking about "what the limitation of NMR is for
accuracy and precision":
How good are your concentrations, pH, tuning/matching, 90 degree pulse, B
zero compensation, temperature control, ... is the receiver gain always the
same or if not have you calibrated it over the range that you are using, dc
offset, what is your signal to noise, how many averages do you need, what is
the field drift, how good is your lock, in the end you are looking at the
area under the curve but how good is your shim.
Hint:
Use an internal standard if possible.
********
Response #2
I would suggest you try deconvolution programs if you use Varian or Bruker
instruments.
I am not sure about other instruments.
The deconvolution program is included in VNMR or UXNMR.
The program will fit the individual lineshape(Lorentzian, Gaussian or mixture
of both) of peaks in the region you are interested.
Hope this info will help you.
**********
Response #3:
Won't go into this in detail; it's an area where a _lot_ of work has been
done over the last 50 years.
1. Baseline correction is likely the single most important factor in removing
common errors in integrals (although see below). NUTS with its FB command
does a very good job of baseline correction. VNMR, Xwinnmr, Winnmr, and other
reasonably sophisticated software do a good job of baseline correction, but
are less straightforward than NUTS to apply. Various sophisticated baseline
correction routines have been written for Felix that go well beyond what is
available in packaged software, but these relate (I believe) primarily to
multidimensional data.
2. Deconvolution fitting is essential to obtaining accurate peak areas when
peaks are not baseline resolved. Again, NUTS and other major software provide
this function. It is important to perform proper phasing and baseline
corrections prior to the deconvolution fitting.
3. Even with the best deconvolution, getting sub-1% quantitation is greatly
improved by 13C decoupling. Other J- or isotope shifted peaks may have to be
accounted for to get the best fitting/areas.
4. Properly accounting for relaxation is essential for good quantitative
work..
**********
Response #4:
I think that the accuracy of most integration is limited by not knowing the
t1. (Work by a collegue:) He could integrate oil samples to within 1% which
was determined by reference to a destructive quantization technique. Most
people seem to get within 10%.************
**********
Response #5:
I recommend the book "NMR Data Processing" by Hoch and Stern (1996,
Wiley-Liss). There is a chapter on quantification, and it deals with the
limitations that you are worried about.
There are some algorithms (curve-fitting) that may improve your numbers, but
the usual integration in combination with a sufficient number of points per
peak (at least 5, better if more), a flat baseline, and excellent
signal-to-noise will do a pretty good job (you probably know that already).
**********
Response #6:
not a controlled answer but I think that the company EUROFINS has deepend the
analysis down to quantifying from the FID to avoid phasing problems possibly
Martin (one of the sons of the creator of this society published his math's)
should appear under chemometrics
**********
Response #7:
We run quantitative analysis for our standards characterization lab. We use
ACD software to process, integrate and create the report. I don't' believe
that the ACD software is any more accurate that the integration routines on
the Varian systems but, it is easier to create a report with
it. The integration routine on the Varian spectrometer is extremely simple,
the system simply sums the intensity of all data points under the region
selected. This is what you want for a quantitative integral. The only problem
is that this is highly baseline dependent. There are some
other, more sophisticated methods for dealing with integration such as Bayes
or deconvolution. Unfortunately, neither one of these methods handles
complex, i.e., coupled, lineshapes very readily. As for accuracy of
integration, we can achieve a routine accuracy of +/- 1%.
************
Response #8:
To ensure the best integrals, make sure that you:
1. Adjust alfa so that lp=0 (using the calfa macro) -- to eliminate the first
source of baseline curvature, then;
2. Perform a two-point back linear prediction (after alfa is properly set)
to eliminate the second source of baseline curvature.
LATER FROM SAME RESPONDER:
If your baseline is FLAT (because alfa is set correctly and the 2nd point
is back LP'ed) you will have no integral drift. (Assuming the phasing is
correct as well.) A "dc" is fine and neccessary, but if you feel the need to
do a "bc" (because of integral drift) you are probably not set up as well as
you could be. Make sure the LP number of points (lpnupts) is twice the
coefficients (lpfilt). To be complete:
parlp - creates LP parameters
dglp - displays them
proc='lp' (vs proc='ft') - turns on LP processing
lpalg='lpfft' - the default
lpopt='b' - for back prediction (vs 'f')
lpfilt=32 - the default
lpnupts=128 - make sure it is > twice the size of lpfilt
strtlp=3
lpext=2
strtext=2
(The last six vales set up two-point back linear prediction.)
To make everything come out well, the phasing, the alfa, and the linear
prediction (all of which interact a bit) need to be set up properly. How you
do this depends a bit on whether you are using any DSP or not. If you are
using DSP, you also need to adjust the rof2 / alfa ratio (keeping the sum
constant) to acheive the flattest baselines.
***********
Response #9:
sounds like you are asking about software solutions.
An interesting way to quantitate without adding internal standard to each
sample is to perform signal injection using ERETIC as described at ENC. In
general, the accuracy and precision of the signal injection method seemed to
be higher than adding an internal standard.
Trouble is, you need a waveform generator on the channel from which you pulse
15N and the probe needs to be one that picks up cross talk between the coils,
like a flow probe. traditional 5mm or 3mm probes will likely not pick it up.
that also unfortuantely means your give up S/N. SO ERETIC presents a
trade-off.
**********
Response #10:
The metabonomics people claimed to have improved everything in getting the
numbers off the spectrum: phasing, etc. This would be Jeremy Nicholson and
John Lindon, aka Metabometrix. And there is another metabonomics company,
Chenomx. (Both have web sites). Software people have probably also thought
long & hard about this; Acorn NMR as a mom & pop operation might be the most
accessible.
There was a paper which reported achieving 1% quantitative accuracy by NMR.
They had to use two technicians, two scales, and rotate between the various
combinations. (Maniara et al, Anal Chem 70:4921-4928 (1998).
The clearest discussion of t1's and tip angles I found was Rabenstein &
Keire's chapter (pp. 323-369) in "Modern NMR Techniques and their
Applications in Chemistry", A. Popov, K. Hallenga, ed.s, Practical
Spectroscopy Series, Marcel Dekker (1990).
***********
Response #11:
In our PERCH Project
http://www.uku.fi/perch.html we have spent a lot of time
on qNMR (quantitative NMR). In our opinion, the "deconvolution method" is the
best tool for integration of NMR signals. For example, in recent experiments
we have found that impurity signals in drugs corresponding to 0.1 mol% can be
easily quantified, if the signals do not overlap strongly. This means that
for small solvent molecules the quantification limit is below 0.01 wt%. We
actively develop qNMR for a number of applications.
Our ten year experience in qHNMR has been incorporated into our PERCH NMR
Software (see the above www-address). The TLS-tool of the software contains
many properties (contraints, powerful base-line functions and options) that
are not found in any other software, so far we know. For an example about the
constraints and multiplet-structures, see Magn.Reson.Medicine,
36:359-365(1996).
Unfortunately there is no demo of the software available at this moment. A
new version, with better Windows-interface, should be available in June. If
you have any questions or like to test the present version please do not
hesitate do contact me again.
***********
Thanks again for your help.
Patrick Hays
Received on Mon May 13 2002 - 13:35:14 MST