Dear All,
Thanks for the great responses. Here is a quick summary and THANKS to all who answered.
Martha
I would recommend using non-linear regression or simplex fitting to a model lineshape with baseline as an explicitly fitted parameter. (this is not necessarily a trival thing to do....). These procedures will also yield statistics on the quality of the fit, though there may be other sources of error.
Baseline issues are bad enough when trying to obtain accurate integrals on 1-D spectra, but they will "bad-squared" on volume integrals.
Do the math: what happens if you include a +/- 5% or 10% baseline error in the volume integral? You will be amazed. With the usual S/N in 2D spectra I would be surprised can know the "baseline" reliable to within even 10%.
Then there all the issues with artifacts, overlapping peaks, etc, etc... (Michael Strain)
You can check with Bruce Johnson (NMRView) and Frank Delagio (NMRPipe). Both programs have forums that have asked that question. It's tougher than one may think. (Ryan McKay)
I used to estimate the error by integrating an area without peaks (only noise) using the same procedure as integrating a real peak. I would get the values from several noise locations, and choose the averaged value as the error for the peak volumes. (Dejian Ma)
If it's like a NOESY experiment, I think the error depends on the range of integration and the S/N similar to a 1D experiment. When I did them, I found the largest cross peak and integrated that with as many times (ideally 5X) the linewidth (from the slices) as possible. Then I used the same integration range for the rest of the cross peaks. David Keire's paper may be of assistance - although it's for 1D quantitation the errors he lists can be extrapolated to the 2D situation based on your S/N and integration ranges (Mark Swanson)
REF: Rabenstein and Keire, "Quantitative Analysis by NMR" in Modern NMR Techniques and Application in Chemistry, 1991
Use big matrix dimensions (SI, fn) to get better numbers. I think the error is on the order of 20%. Just an opinion. (Jeff Simpson)
With 60 spectra over 60 hours, the volumes from NMRView have around an R^2 value of around 0.25. FIDRES for the N-HSQC experiments used are about 7 Hz in the direct dimension and 24 Hz in the indirect. Many of the correlations exchange during the experiment. We were concerned about the data quality, but not anymore. Thanks to all!
Martha Morton, PhD
Director of Research Instrumentation
University of Nebraska - Lincoln
Department of Chemistry
834 Hamilton Hall
Lincoln, NE 68588-0304
402-472-6255
Received on Fri Feb 21 2014 - 05:03:48 MST