3D training

Here is a full-immersion course for you to learn 3-D processing. The material consists into iNMR (be sure to downlaod the most recent version; an unregistered copy of iNMR reader is enough) and the following spectra:

Anybody can contribute with more material. It's enough to send the link to a Bruker or Varian dataset and the permission to put it into our collection.


When you double click the iNMR document, all the processing is automatically performed. What you see is a single plane of the transformed cube. To move to the next plane, press the key t. Press r to navigate in the opposite direction. To rotate the cube, choose Format > Axes & Scales, go to the last panel of the dialog and choose which dimension to see along X and which along Y.

Processing History

Too examine how a spectrum has been processed, choose Edit > Copy > Processing, then paste the summary anywhere in order to read it. The basic workflow is:

The phase correction parameters are not listed into the summary. To examine them, reload the FID with the command File > Reload. (If you turn on the preference “Show the FID when Opening a Document” you can always start from the FID but you can't use the command Copy Processing). If the summary starts saying that a signal has been suppressed, open the dialog Process > Suppress and click OK. In the same way, open all the dialogs in the above list. This is how the spectrum was processed. Being familiar with the graphic representations of the spectrum during all the stages of processing is an important skill that will help you when you are going to process your own spectra.

Parameters to Ignore

The FT dialog contains many options that are not relevant in the field of 3-D NMR: Real FT, Magnitude, Subtract D.C. are options you should never consider. You could use the options: Live, Use Linear Prediction, Exponential, Gaussian, “Cut After”, but you can also process all your 3-D spectra perfectly without ever using them.

Compulsory Parameters

These parameters are automatically set by iNMR. You can skip this paragraph on your first reading.

There are 3 parameters that always have a dramatic effect on any FT of any dimensionality. In each case there will be a unique combination of these that will produce meaningful results and no other combination can be used. The effect of the first two options is what their names implies: “Swap Sides” and “Mirror Image”. The third parameter (shuffling) should never be used along the directly acquired dimension (the first FT). When a nD spectrum is “Phase Sensitive” you must use the shuffling with the second and the third FT. All our examples are phase sensitive.

Triple resonance experiments are HSQC-like experiments that require a special kind of shuffling, called “echo-anti echo”. Varian spectra obey strange rules. While 2-D HSQC Varian spectra require the echo-antiecho shuffling at the moment of FT, 3-D spectra require this operation when the FID is loaded from disc. This kind of processing is performed before you can even see the FID on the screen. You must use simple phase-sensitive shuffling in the second and third FT. You can disable the pre-processing with the command rkpp(false) if iNMR applies it when it should not, for example in spectra like hcchtoc of our examples. To examine the current pre-processing status: print(rkpp()).

Troubleless Parameters

The size of the FT is important but not critical. A simple rule that always works great is to double the size of the FID. Remember, however, that shuffling halves the size of the interferogram. The rule of thumb, for phase-sensitive experiments, is: the size must be at least the double of the number of points acquired in the direct dimension and at least the number of points in the other dimensions. “At least” means: the number itself or the next power of 2.

Weighting is the same: important but not critical. You can use a squared cosine bell for your phase-sensitive experiments. In the direct dimension you can also use a simple cosine bell.

Strategic Parameters

There are essential operations for which a general advise can't be given in advance. These operations are: solvent suppression, first-point-pre-multiplciation, phase correction and baseline correction. Note that in many of our examples one or more of these treatments are not applied and they should not. The solvent is normally suppressed when the transmitter (center of the spectrum) is exactly on the water signal, not in other cases. The baseline correction is rarely used along all the dimensions. Often the phase is corrected along the direct dimension only, and only using the zero-order correction. All these treatments are connected and influence one another. With a little of experience you can predict what you need to change after seeing the transformed spectrum.

First Point Pre-Multiplication

Sometimes you can see, in the final spectrum, that the columns (or rows) containing a peak are monochromatic. The corresponding traces have a flat baseline. Flat and horizontal, but non-zero. A zero-order polynomial correction can remove this defect, more in theory than in practice. Its weakness is the automatic sampling that is never perfect. A common result is an excessive correction (the lines changes color, but remain there).
An equivalent correction, but performed in time domain, is the multiplication of the first point of the FID with a suitable coefficient. In the ideal case (no delay before the first point of the FID) this coefficient is 0.5. Many 3-D experiments are acquired to satisfy this ideal condition.
A simple recipe for you to follow is to try, at first, to multiply the first point (in all dimensions) by 0.5 and to avoid using the first-order phase corrections. Observe the result. If the phase is never acceptable introduce the first-order correction in that dimension only. If the baseline is ugly, renounce to the pre-multiplication (in that dimension only) and reprocess.

Phase Correction

Extract the first row from the FID, process and correct the phase (only the zero-order correction). Go on with the 3 FTs on the whole matrix, using the hypercomplex option along the indirect dimensions. Transpose the cube to display the f2-f3 planes. Choose a plane with many signals, possibly not concentrated in a single region. Correct the phase for all the visible signals. Verify that all the other parallel planes are OK too. Transpose to display the f1-f3 planes and repeat. Now everything is correct. Sometimes things can be more complicated than this, sometimes they can be simpler.

Baseline Correction

There are 3 possible directions for the correction: X, Y and Z. If corrections are required in more than one dimension, they should be performed with a single command and in order, from the most needed to the less needed. The indirect dimensions do not contain enough points for fitting a polynomial to them. Fortunately they often don't need a baseline correction, but when it is necessary, prefer the smoother (uncheck the option “fit with a polynomial”). This kind of correction is also effective in removing broad signals (like water). Verify the effect on a selected trace before applying the correction to the whole matrix.

Extracting a Sub-Matrix

Cutting away half of the spectrum obviously cuts down memory requirements and computing times but complicates baseline correction, because the latter requires a lot of experimental points and empty regions at the bondaries of the spectrum. If your spectrum is clean and does not require any baseline correction, you can certainly export a submatrix and work with it. In all cases, you can export a submatrix when you have arrived at the end of processing, after baseline correction.

Linear Prediction

None of our examples employs LP, mainly because we wanted to give you examples that you could reprocess in a few seconds. You are invited to try implementing LP, instead of zero-filling, but only along the indirect dimensions, never along the direct dimension. First, however, process a spectrum, the best that you can, without LP, then compare the results obtainable with LP. It is a weapon, but not always beneficial. There are 3 algorithms for LP. The most convenient one is Singule Value Decomposition. The Fast LP (equivalent to the LP-filling option of the FT dialog) can often yield similar results in much less time but other times it is numerically unstable.

Visual Guide

Before reading this page again, study the illustrated tutorial. Within 5 minutes, the same concepts will become easier to understand.

Graphic Tricks

A proper setting of the contour levels can greatly enhance the resolution, the sensitivity and the readability. Set the treshold high and keep the levels near to one another. Limit the number of levels: peaks are easier to recognize when the top appears flat. If all the peaks are positive, hide the negative levels.

Your Training

Pick one of our examples and change the name of the file plist.inmr. Now open one of the remaining files with iNMR and try processing the spectrum by yourself. If you get lost, delete the new plist.inmr, rename the old file correctly and pay attention to our parameters.
A less challenging training consists in duplicating the whole data set and process the two copies in alternation, step-by-step. Keep the plist.inmr file into only one of the copies, so you will be forced to manually set the parameters into the other one.

Copyright © 2005-2021 nucleomatica
Valid XHTML and CSS. UTF-8 encoding.