Wednesday, 21 October 2009

Binning and NMR Data Analysis

Yesterday I mentioned that many NMR arrayed experiments suffer from unwanted chemical shift variations due to fluctuations in experimental conditions such as sample temperature, pH, ionic strength, etc. This phenomenon is very common in NMR spectra of e.g. biofluids (metabonomics/metabolomics) but also exists in many other experiments such us Relaxation, Kinetics and PFG NMR spectra (diffusion).
This problem negatively affects the reliability of quantitation using, for instance, peak heights, and for this reason integration is, in general, a more robust procedure as these spectral variations are mitigated by averaging data points over the integral segment. In this post, I just want to show you one simple trick which helps to understand, in a pictorial way, why integration is useful to remove the major part of chemical shift scattering.
First, consider the following experiment depicted in the figure below. It shows a triplet and as you can see, some minor peaks shifts are present from spectrum to spectrum

If peak heights are determined at a fixed position, this might introduce appreciable errors in the posteriori quantitative analysis (e.g. exponential fitting). As described in my former post, this could be circumvented in some extent by using parabolic interpolation or peak searching of the maximum in a predefined box.
Nevertheless, integration is a very simple solution as can be appreciated in the figure below. Instead of using the Peak Integrals tool in the Data Analysis module, I will show now a complementary procedure. Basically, what I have applied to all spectra is the well-known binning operation which consists of dividing each spectrum in equally sized (e.g. 0.01 ppm in this case) bins, so that integral (area) of each bin represents a new point in the binned spectrum

As seen in the figure above, binning clearly removes the effect of chemical shift changes but of course, at the cost of a significant reduction in data resolution.


stan said...

Hi Carlos,
as you probably know, I am no fan of the concept of binning. It has several weak points, the most severe being that when a line is close to the edge of a bin, it gets split artificially in two sections which end up in different bins. As your own figure above shows, the intensity errors in the different bins are not much better than when doing no binning at all.
Of course, I agree that Mnova (or any other NMR software) should offer the binning for completeness, though as a User, I would hardly ever use it. For most purposes, a suitable Gaussian apodization does the same trick, for example.
Years ago I have used a trick which did essentially the same service as binning, but avoiding some of its drawbacks. It is just an apodization with a suitable sinc function before FFT which, I think, you definitely should include in your list of apodization options. I have used it in my old DOS software in the 80's (we did not know the term 'binning' then) and it worked like charm.
Of course, I did not publish it; you know how it is (there are three types of people: those who do, those who teach, and those who publish). But if you get in touch, you can implement it in 2 hours and try it on the same data.
Ciao, Stan

Carlos Cobas said...

Hi Stan,
My only intention with this post was to show, graphically, the effect of using the ‘Peak Integrals’ option in the data analysis module for the extraction of peak intensities from arrayed NMR spectra. I reckon that the integration method (which is equivalent to the combination of binning + Peak Heights retrieval) has many limitations (I mentioned some of them in one of my previous posts – e.g. it is not reliable whenever there is any peak overlap)
In particular, regarding binning, there are some ‘intelligent’ binning methods including:
* ‘Adaptive binning: An improved binning method for metabolomics data using the undecimated wavelet transform’ (Chemometrics and Intelligent Laboratory Systems 85 (2007) 144–154)
* ‘NMR-Based Characterization of Metabolic Alterations in Hypertension Using an Adaptive, Intelligent Binning Algorithm’ (Anal. Chem. 2008, 80, 3783–3790)
* ‘Robust Algorithms for Automated Chemical Shift Calibration of 1D 1H NMR Spectra of Blood Serum’ (Anal Chem. 2008 Sep 15;80(18):7158-62)
As you know, I have implemented an automatic alignment algorithm for spectra which have been previously deconvolved using GSD but so far I have used it only with PFG experiments.

Thanks for your comment, Stan.