Frank Delaglio NMR Science Home - 1D and 2D NMR series Analysis, NMR Screening

Welcome: an Introduction to this Site

NMRPipe: a Comprehensive Software System for Biomolecular NMR

Special Applications: NMR Drug Screening, Automation, and so on

Dipolar Couplings, Chemical Shifts, and Protein Structure

Miscellaneous

Frank's CV, Recent Slides, and Thesis

  Frank Delaglio, Ph.D.

  Software Science Consultant
  13840 Grey Colt Drive
  North Potomac MD 20878 USA

Tel: 301 806-0867
  Fax: 301 309-8717
  E-mail: delaglio@nmrscience.com

Special Applications: NMR Drug Screening, Automation, and so on

Scan File System for Spectral Data, Automated 1D Batch Processing and STD Analysis

The selNMR graphical interface automatically scans directories for spectral data, and extracts and displays the spectral parameters so that data can be selected on the basis of experiment type, size, dimensionality, etc. Spectral data selected by selNMR can then be directed to other applications such as ezProc1D for selection of batch processing parameters, with automated phase correction and PostScript plot generation. There are also specific tools for automated 1D STD (Saturation Transfer Difference) analysis, including adaptive rescaling, and generation of STD PostScript plots which summarize the STD values for each peak and for the overall spectrum.

2D HSQC Screening, Multivariate Analysis, and Chemical Shift Titration

Many applications of 1D and 2D NMR spectral series analysis have found use in the drug design and evaluation process. In particular, the elegant approach of Fesik and coworkers to use protein 2D HSQC perturbation screening as part of a comprehensive scheme for drug design is now well known (Shuker et al., 1996). NMRPipe facilities have been used to create special applications to address a variety of HSQC analysis tasks. Some of the applications are suitable for any collection of related HSQC spectra acquired with similar parameters, and can be used with collections of several spectra or several hundred. Other applications are intended primarily for HSQC chemical shift titration series.

Automated Fitting of Chemical Shift Evolution

Interactive Fitting of Chemical Shift Evolution

HSQC Series Analysis: Spectra of Aggregated Samples (red)

HSQC Series Analysis: Spectral with pH Effects (magenta)

HSQC Series Analysis: Spectra with Additional Scans (green)

HSQC Analysis: Spectra with Specific Binding Mode (blue)

The AutoProc application automatically processes a related series of HSQC spectra, with automatic phase correction of the directly-detected dimension. It uses as input a list of the spectra to be processed, and a representative NMRPipe conversion script for any one of the individual spectra in the series.

The TitrView application follows the change in position of one or more peaks in a spectral series. The results are summarized in a single table. As such, TitrView is a complement to PCAView, which does not use peak table information. Facilities are provided for automated peak picking of the series, automated tracking of each peak's position, and interactive adjustment of the results, to insert, move or delete peaks. In the TitrView graphical interface, the initial results are found automatically, and confirmed and adjusted interactively. In the display, a given row follows the evolution of a peak's position over the series, as indicated by cursor lines; the first entry in each row shows the given region drawn in overlay for all spectra in the series.

The ModelTitr application is intended specifically for ligand titration series. It provides a method for estimating dissociation constants (Kd) for ligand binding to individual residues in the target protein according to the 1H or 15N chemical shift evolutions in the HSQC titration series (Zhou et al., 1996; Johnson et al., 1996). ModelTitr fits each peak position evolution curve to estimate a dissociation constant Kd for the corresponding residue. The application uses the results of TitrView as its input, and works non-interactively. The results for each curve are summarized in PostScript plots. Either HN, 15N or a weighted combination of the two types of shifts can be analyzed. The ShowTitr application is an on-screen interactive alternative to ModelTitr. The ShowTitr graphical interface can step through and display the individual evolution curves, and apply fitting routines to selected curves via the script fitXY.tcl. The application uses the results of TitrView as its input. As with modelTitr, either HN, 15N, or a weighted combination of the two types of shifts can be analyzed.

The PCAView application is a unique approach to spectral series analysis that allows an entire spectral series to be summarized and evaluated graphically, without the need for peak picking. Instead, the complete matrix of intensities for each spectrum is used directly, and spectra are clustered according to how similar their overall collection of intensities are. The clustering is performed interactively by inspecting the results of Principal Component Analysis (PCA). As a qualitative screening technique, this method has been used to highlight cases where spectral perturbations are due to nonspecific effects such as pH, to reveal cases where meaningful changes to spectra have occurred, and to identify different modes of binding. (Ross et al., 1999). The PCA technique can also find use in other types of spectral series analysis, including 1D metabolite screening of biofluids. Because the PCA method does not require peak analysis, it can be quick and effective even in cases where the HSQC spectra are typically not fully resolved.

Multivariate Representations and Principal Component Analysis

The PCAView application makes use of a direct multivariate approach, where an entire spectrum is represented as a single object in a multidimensional space. The coordinates of that object are simply the intensities at each point of the spectrum. There are some useful properties of this representation. For example:

Similar spectra will cluster together in the same region of the multidimensional space, since their intensities (i.e. their multivariate coordinates) are similar.
Spectra with similar features but differing intensity will cluster along lines and curves in the multidimensional space. This is because they will have some intensities in common, and some others which vary continuously.
Weaker spectra will cluster nearer to the origin of the multivariate space than more intense spectra.

The multivariate space can be visualized by projecting it onto a smaller number of dimensions, by the method of Principal Component Analysis (PCA). A given principal component points in the direction of maximum variance in the data. The PCAView application implements the method of Ross and coworkers for analyzing HSQC drug screening series (Shuker et. al., 1996) by Principal Component Analysis (PCA) (Ross et al., 2000). The application uses multivariate statistics to provide a graphical summary of the similarities and differences in a collection of related spectra, in this case a series of automatically processed HSQC spectra of roughly 200 samples of a target protein mixed with various small molecules. Each number in the scatter plot at lower left of the PCAView graphical interface represents an entire HSQC spectrum in the series. The distance between entries in the scatter plot relates to the degree of similarity between spectra. The spectral window on the right of the graphical interface allows one or more spectra or regions from the series to be viewed in overlay.

In the examples shown above, the classes have already been interactively selected and colored after inspection. The yellow cluster reveals the bulk of spectra, which are mostly unchanged. The green cluster is a subgroup of spectra, which were acquired with experimental conditions which make them more intense than the others. The spectra in the red cluster are all exceptionally weak; the protein in these samples has aggregated. The magenta cluster reveals samples that have undergone extensive pH changes, resulting in systematic elimination of certain peaks. The blue cluster reveals spectra with collections of peaks which have moved systematically, indicating that binding has occurred.

References

Johnson, P.E., Tomme, P., Joshi, M.D., and McIntosh, L.P., (1996) Biochemistry, 35, 13895-13906.
Ross, A., Schlotterbeck, G., Klaus, W., and Senn, H. (2000) J. Biomol. NMR, 16, 139-146.
Shuker, S.B, Hajduk, P.J., Meadows, R.P., and Fesik, S.W. (1996) Science, 274, 1531-1534.
Zhou M., Harlan J.E., Wade, W.S., Crosby, S., Ravichandran, K.S., Burakoof, S.J. and Fesik, S,W. (1996) J. Biol. Chem., 271, 31119-31123.

A Prototype Application: Measuring Distances and Angles on Medical Images

This illustrates a proof-of-concept application for distance and angle measurement over a series of related medical images. In this case, Hip-Knee Angles (HKA) which were previously measured on one large piece of conventional X-ray film, now had to be measured accurately over a series of three smaller digital X-ray plates. The specifications also anticipated use of graphical templates for fitting of prosthesis. Image data courtesy of Dr. Hans Grahn.

A Custom Application: Display and Multivariate Analysis of 3D Spectral Image Data

In this example, each pixel of the image consists of a 1D spectrum; in this case an infrared (IR) spectrum of dye-labeled cells taken with a Fourier-transform IR microscope. A conventional microscopic image at a single given frequency is unrevealing, however a principal component composite image clearly differentiates between types of cells. This graphical interface was part of an overall application for automated conversion, Fourier transform, and background correction of IR spectral image data.

[ Home ] [ Terms of Use ] [ Copyright ]
last updated: July 17, 2008 / big frank

Contact site designer mimi at mdelaglio@hotmail.com