Spectral Processing & Analysis - Applications & Techniques in NMR Spectroscopy
Scientists at the NIH and J.W. Goethe University, Germany have used FELIX for spectral processing and analysis of NMR data.

Improved Spectral Resolution Using Linear Prediction
Frank Delaglio, National Institutes of Health, Bethesda, MD, USA.

Easy, Fast Extraction of Coupling Constants from Multidimensional Spectra
Prof. C. Griesinger and A. Rexroth, Institute of Organic Chemistry, J.W. Goethe University, Frankfurt am Main, Germany.

ND Peak Fitting for More Accurate Spectral Data Extraction


-----------------------------------------------------------------------------------------------------------------------------------------

Improved Spectral Resolution Using Linear Prediction
 

21kb

In the right panel, a 64 data point FID has been zero-filled to 128 data points and Fourier transformed into its 1D spectrum. In the left panel, the same data were extended to 128 data points using the enhanced linear prediction capabilities within FELIX. The linear prediction clearly improves the resolution of the resulting spectrum.

Fourier transformation is an indispensable tool for converting time-domain data into interpretable spectra, although its limitations in reconstructing spectra from small numbers of measured points have been highlighted recently in multidimensional spectroscopy. The usual Fourier processing requires artificial dampening of the measured data with a window function (apodization), followed by extension of the data by zero filling. In a sense, this procedure discards precious data since the original signals are attenuated. In the case of data with many points, this loss can usually be tolerated. But in the case of data with few points, Fourier processing leads to excessively broad spectral lines, as well as extreme truncation artifacts, which make spectra difficult to analyze.

The FELIX implementation of linear prediction solves this problem by extending the measured data with synthetic points before Fourier processing (Figure 1 -21kb). By extending the data, the line-broadening caused by apodization is reduced, and truncation artifacts are minimized. The FELIX macro approach to spectral processing is especially helpful since it allows dimensions to be processed or reprocessed conveniently in any order. The FELIX implementation of linear prediction also includes facilities for mirror-image linear prediction(1), a clever method designed to exploit the symmetry of NMR time-domain data in order to predict larger numbers of signals accurately from a limited number of data points. Together, these features and enhancements make FELIX an excellent solution to the challenge of multidimensional spectral processing.

Contributed by Frank Delaglio. Frank Delaglio is a software scientist at the National Institutes of Health in the lab of Dr. Ad Bax. He has been a consultant of Accelrys since 1991.


-----------------------------------------------------------------------------------------------------------------------------------------

Easy, Fast Extraction of Coupling Constants from Multidimensional Spectra

17kb       38kb

When determining the structure of proteins, torsion angle information resulting from NMR spectra can tremendously speed-up the convergence rate of structure calculations and provide more precise structures. Recently, a method(2) was designed to measure with high sensitivity 3J(HN,C') coupling constants in proteins that are associated with the backbone torsional angle φ. Three 1H,15N-HSQC spectra are recorded displaying the couplings to the carbonyl resonances: (a) in an E.COSY manner, (b) only during the detection of the HN resonance, and (c) not at all (Figure 2 -17kb). The desired 3J(HN,C') coupling constant can be extracted from these three spectra for each crosspeak by determining the 2J(HN,C') coupling, first from the E.COSY type experiment, then with this knowledge fitting the HSQC (c) by convolution with the already determined 2J(HN,C') coupling and the desired 3J(HN,C') coupling to the HSQC (b). The implementation of this procedure for each crosspeak in a set of three 2D HSQC's or 3D HNCO spectra is done by manual or automatic peak picking of all crosspeaks in the spectra and storing them in the database. If an assignment already exists, only the E.COSY type crosspeaks have to be picked again, since two regions have to be defined per individual crosspeak. The flexible FELIX macros provide all the 3J(HN,C') coupling constants in a fully automatic, reliable way. The easily-extended 3D version of the same macro offers the opportunity for analyzing 3D HNCO spectra in cases of severe overlap. The output consists of the two coupling constants of interest. In the graphic window, the reference and the fitted traces appear, as well as the difference between the experimental trace and the trace obtained by the best fit (Figure 3 -38kb).

Contributed by Prof. C. Griesinger and A. Rexroth, Institute of Organic Chemistry, J.W. Goethe University, Frankfurt am Main, Germany.


-----------------------------------------------------------------------------------------------------------------------------------------

ND Peak Fitting for More Accurate Spectral Data Extraction
 

49kb

Peak fitting is critical for the accurate interpretation of NMR spectra. Accurate peak centers are required for computer-assisted resonance assignment, especially in higher dimension spectra where the digital resolution is low. The calculation of many spectral properties, such as relaxation and NOE build-up rates, are dependent on the accurate measurement of peak linewidths, intensities and volumes.

FELIX contains extensive optimization procedures for fitting 1D, 2D, 3D and 4D peaks. The centers, widths and volumes can be optimized independently or simultaneously in any combination or order with Gaussian or Lorentzian lineshapes using a linear least squares fitting routine. The residual volumes can be visualized quickly to verify the accuracy of the fit. Frames 1-3 of Figure 4 (49kb) depict the experimental, model and residual crosspeaks within a small region of a 2D NOESY spectrum of Zinc Rubredoxin. Frames 4-6 depict the experimental, model, and residual crosspeaks after optimization of the peak centers, peak widths, and volumes was conducted. Note that both positive (red) and negative (green) contours are plotted in the residual data (frames 3 and 6) and that very little residual intensity is left after fitting of the peak footprint (frame 6).


-----------------------------------------------------------------------------------------------------------------------------------------

Back to top
More Case Studies