Field Of Genes

In a recent issue of The Scientist, I reviewed, in a very general way, the status of data mining and presentation software with specific reference to gene expression arrays. In the course of that effort it became clear to me that detailed reviews of some of the more well-known and widely used of these software packages would be useful. This is the first in a series of articles designed to explore these packages. It is my intention to use each of the applications reviewed in an analysis and presentation of expression array data. Mathcad products have been in production since 1984 and currently are in wide use among engineers, physicists, and to a lesser extent chemists. Company estimates put their market share at 30-40% of the potential users in these disciplines. Two features of Mathcad 8 are easy to appreciate immediately. The first is its total integrated compatibility with Windows NT/95/98. This is not currently the rule among the high-end mathematics packages. This compatibility and the ease with which large data sets can be handled as vectors or matrices by the nonexpert should make Mathcad 8 attractive to biologists in a number of areas besides functional genomics. Second, the help resources are multileveled and clear. Help is available from a variety of sources: a resource center, online help, a collaboratory, quick sheets, a user chat room, and traditional technical support. In addition, electronic books and extension packs are available from Mathcad for a variety of applications. Excerpts from some of these appear in the resource center. Figure 1: A surface plot in which increasing message expression levels are indicated in the color change from blue to orange along the Z axis. Time in the cell cycle and gene number are in the X,Y plane. Figure can be inverted to show minimum expression levels in blue. Data taken from Cho et al. as available at genome-www.stanford.edu/cellcycle I found the resource center and the quick sheets most helpful. If your problem is similar to one that others have dealt with, a preformed solution is likely to be available and can be brought into your own Mathcad worksheet, often with a simple click and drag. In general, I found that even low-level data handling features were superior to those in Excel, and functions that are possible but arcane or awkward in Excel are rather simple in Mathcad. Transposing rows for columns in an entire file is done using standard mathematical notation and can be invoked via the math palette or simply by typing Ctrl-1. In Mathcad it is possible to convert an Excel file to a color-mapped or black-and-white image, to unwrap the file into a vector, to import it and maintain its internal structure as an array--all with a few clicks of your mouse. A passing familiarity with vector and matrix math will probably speed your climb up the learning curve, but for many simple operations, even this is not necessary. Two- and three-dimensional graphing yields some rather dramatic images that remain "live" in the worksheet so that multiple viewpoints or recalculations are instantaneous. As someone who never gets a calculation right the first time, I particularly appreciate the instantaneous recalculation feature. Did you leave out an essential normalization early in a lengthy calculation? No problem. Insert it at the appropriate spot in the worksheet and the entire process and the graph derived from the calculation are instantly redone. Prior to receiving a copy of Mathcad 8, I had never used this software. After a week or so of fumbling and posing a dumb question or two to the technical support desk, I was able to take large 6,000-20,000 row Excel files; transpose and represent them as color-mapped images, surface plots, contour plots, or vector field plots; or unwrap and create a vector, which could then be analyzed by wavelet or fast Fourier transform (fft) functions for time structure. Starting with an Excel file containing the results of the Cho et al.1 study of gene expression through the yeast cell cycle, it was possible with just a few clicks and keystrokes to create color or gray-scale images of the entire cell cycle for all 6,178 genes (Fig.1). Once these files have been read as images, it is possible to apply the whole spectrum of image processing tools to the images to discover large-scale structures or textural features. Figure 2: Cell cycle Phase and Message expression levels for three genes through two cell cycles in Yeast. Increasing concentration is indicated using the same color map as Figure 1 Some of us like to think about the cell cycle as an attractor: a preferred path through message concentration space in which time (cell cycle time) is implicit in the length of the closed path. If you share this deviance, Mathcad will let you plot these relationships rather quickly for any set of genes you wish. Here, I chose three that when plotted together resemble the Lorentz attractor (Fig. 2). Showing the relationship in time and message concentration between any subset of genes typically begins by transposing the Excel file; this is followed by extracting and giving appropriate variable names to the desired set of genes and then representing them in 3-D plots of relative changes in concentration of individual messages through two cell cycles. It is then just a matter of identifying the columns of interest as variables and then entering them into the place holders at the bottom of the figure. Regions in Mathcad worksheets are live, and figures can be grabbed and rotated. With some additional effort, it is possible to animate the data to reveal features that are not apparent in single still figures. Extension packs for signal processing, image processing, and wavelets are also available, and I used the imaging and signal processing features in the work described here. However, serious time series analysis would probably require the companion offering from MathSoft: S-Plus. The worksheets are designed to function seamlessly to calculate or graph, to describe the results of these calculations in the word processor, and to then disseminate the results to others via the Internet. The worksheets force standard (and therefore correct) math notations on the writer, and errors are color coded and elaborated as one proceeds through a document or calculation. Mathcad can create linked documents, electronic books with live math, and hyperlinked worksheets for presentations and instruction. The hyperlink feature of Mathcad allows you to present a series of files as a single unit, similar to a Mathcad Electronic Book. You can also add small popup windows and link to documents on the Internet or applications outside of Mathcad through the components and MathConnex. Users familiar with visual programming environments will be immediately comfortable in MathConnex (available only in Mathcad 8 Professional). MathConnex allows you to automatically run data or calculations through a variety of other applications. Mathcad uses the Microsoft ActiveX scripting languages--VBScript (Visual Basic Scripting Edition) and JScript (Microsoft's implementation of JavaScript)--to allow any OLE automation supporting object to be used in a Mathcad computations workbook. Axum, Excel, File Read/Write, Data Tables, Matlab, and S-Plus are all directly connected without the need to write a script. Otherwise, a generic connection encodes the Mathcad component but requires you to write code in the target program's language. For improved data mining, one might wish that neural network and fuzzy logic extension packs were available. Mathcad has just released a wavelets extension pack that was not part of the applications group I had for this review. That extension pack appears to have a rather complete set of the types of signal processing and time series power that was previously found only in S-Plus. For the experienced programmer, Mathcad may not be as suitable for numerical simulation programming as other products. However, compared with others in its class, Mathcad is a very economical solution to our need for sophisticated data analysis and graphing software. One caution: Mathcad will run on a 486 machine with 16 MB of RAM, in theory. To avoid endless waiting and page swapping, however, problems of the sort discussed here require 128 MB of RAM and a fast machine. This product should be of interest to individuals, groups or core facilities in bioinformatics, functional genomics, image processing, and cellular dynamics. R.J. Cho et al., "A genome-wide transcriptional analysis of the mitotic cell cycle," Molecular Cell, 2:65-73, 1998. For more information, please contact MathSoft at 800-628-4223, or visit the company Web site at www.mathsoft.com. The author can be contacted at neumann@earthlink.net.

Field Of Genes

Interested in reading more?

Become a Member of