Test Platforms

G3 Power Mac 300 MHz, 160 Mb RAM, running OS 8.1 (Prism 2.0b)
Pentium 150 MHz, 80 Mb RAM, running Windows 95 (Prism 2.01)

Overview

GraphPad Prism combines biostatistical analysis, curve fitting, and scientific graphics into a single, user-friendly, intuitive package. Unlike many other statistics programs, Prism specifically addresses the types of analyses commonly utilized in biomedical laboratories. Data is handled in a project format, which allows data, graphs, data analysis, experimental notes, and presentational layouts to be saved in one file. Users can customize nonlinear regression analysis by entering their own equations; there are also nineteen built-in equations that cover most curve-fitting needs. A notable feature is the built-in F-test, which is used to compare two models statistically to a data set. Presentation graphics are of publication quality, and graph attributes are fully customizable. Furthermore, data and results tables may be easily combined with graphical results on the same page using one of a large number of predefined layouts.

For those less confident in statistics, the well-written manual provides guidance about not only how to choose appropriate statistical methods, but how to interpret the results. A well-designed Web site maintained by company president and program designer Harvey Motulsky (a pharmacologist with a background in biostatistics) offers program technical support along with a wide array of helpful biostatistical aides and resources. These resources further guide users toward the proper use and interpretation of the statistical analyses included in Prism. Prism is an ideal package for students and researchers alike.

Program Features

Figure 1

Prism offers users an all-in-one package for biostatistical analysis, curve fitting, and scientific presentation. All pertinent information relating to a data set is linked together and saved as one file, which enables Prism to calculate and update graphs automatically as new data is entered or modified. Replicate data sets are automatically graphed as the mean value, with the standard error of the mean (SEM) shown as error bars. If preprocessing of data is desired, Prism offers the capacity to transform, normalize, prune large data sets, subtract baseline values, or transpose rows and columns (figure 1). Many built-in transforms are provided, along with the option to add user-defined functions.

Project Management

Figure 2

Upon opening Prism, the user is greeted by a welcome window from which he or she can start a new project, open an existing project, or open one of the last nine projects recently worked on (figure 2). Each project is composed of five sections - data, results, graphs, layout, and notes - and may contain up to 100 sheets per section. All of the sections in a project are stored as one file so that data, analyses, graphs, and notes for an experiment are kept together. It is possible to merge or split projects as more data is collected. One of Prism's strengths is that it automatically calculates mean and standard deviation, graphs data, adds error bars to graphs when replicate data is entered, and updates calculations without intervention by the user as data is added to the worksheet.

Tools and Templates

Figure 3

Prism has three levels of tools contained on the toolbar at the top of the page. The top-row buttons control actions relating to global attributes of the project, such as creating new projects and opening, saving, and printing documents. The second row consists of navigation tools specific to the five project sections: data, results, graphs, layouts, and notes. The third-row tools change depending upon which of the five sections is in the active window (figure 3).

There are nine template files designed to give users a sense of how to analyze data using Prism. These templates cover frequently used analyses for creation of histograms, analysis of radioimmunoassay data, nonlinear regression to analyze saturation binding data to create Scatchard plots, simple graphing, linear regression to generate standard curves, nonlinear regression to produce summary tables, unpaired t-tests, analysis of enzyme kinetics, and a more advanced analysis of radioligand binding data. The Notes section of each template contains a brief description of how to use it and what type of analyses it will automatically perform. New templates may be created, and all existing templates may be modified to fit individual users' needs.

Importing and Exporting Data and Graphics

Prism can import data saved as tab-, comma-, or space-delimited ASCII or plain-text files. Currently, dynamic data exchange (DDE) format and object linking and embedding (OLE) of graphics are not yet supported. Data files in ASCII or plain-text format can be generated easily from most spreadsheet programs. Graphs and layouts can be exported either by copying and pasting from the clipboard or by exporting the graph as a WMF, BMP, PCX, TIF, or GIF file for the Windows version of Prism, or as a PICT file for the Macintosh version.

Statistical Analyses

Prism covers all the basic statistical analyses used by biomedical researchers. These include parametric and nonparametric t-tests, one- and two-way ANOVA, correlation, linear and nonlinear regression, and survival analysis. Other analysis types that are included allow data smoothing using the method of Savitsky and Golay, and calculation of the area under a curve. Summary statistics such as mean, standard error, and standard deviation are automatically calculated from replicate data, and error bars are automatically added to graphs.

Linear and Nonlinear Curve Fitting

Standard curves are frequently generated in the laboratory to calibrate an assay. One determines an unknown parameter, such as the concentration of a substance, based on a measurable parameter such as absorbance, radioactivity, or luminescence. In order to obtain values for the parameter of interest for the experimental samples, a line or curve must first be fitted through points generated using known concentration standards within the range of the experimental analysis. This may be done using linear regression, nonlinear regression, cubic spline, or polynomial regression. All that matters for standard curve fitting is that the curve must fit the standard points as closely as possible. Other applications for nonlinear curve fitting require that the equation be chosen very carefully.

Figure 4

Once the standard curve has been generated, measurements from experimental samples are compared to it and interpolated values are obtained for the unknowns.

Figure 5

Figure 6

Figure 7

Linear regression may be used to generate a standard curve if experiments are designed to stay within the linear region of the curve. In the case of a protein concentration standard curve, one would enter the known concentrations and the corresponding absorbance measurements (figure 4); Prism will then automatically graph the data points (figure 5). The parameters for curve fitting, however, must be chosen by the user. This is done by selecting the Analyze button on the toolbar while in the data window (figure 6). The graph will now show how well the data points fit to a straight line. To read unknowns from this curve, one simply adds the absorbance values of the unknowns to the same data set beneath the Y values of the standard curve. To view the automatically interpolated values, one switches to the results section and then, from the View menu under Select View, chooses Standard Curve X from Y (figure 7).

Figure 8

For nonlinear regression curve fits, Prism calculates the fits by minimizing the sum of the squares of the vertical distances of the data points from the curve. This calculation method can be weighted or nonweighted; users should decide this depending upon where experimental error may be contributing to the scatter of the data. Many options are given in the Parameters dialog window (figure 8). To aid in analysis, Prism automatically provides initial values for all variables if the user does not supply them.

Figure 9

Prism includes nineteen built-in equations for nonlinear regression curve fitting tailored to the analysis of biochemical or pharmacological data. These functions include one- and two-site binding, sigmoidal dose-response (both constant and variable slope), one- and two-site competition, Boltzmann sigmoid, one- and two-phase exponential decay, one- and two-phase exponential association, exponential growth, power series, polynomial equations, and sine-wave and Gaussian distributions.

Figure 10

In addition to the built-in equations, user-specified equations may be entered (figure 9). Prism can automatically perform an F test to compare the fit of two models. One selects this option in the Parameters dialog for nonlinear regression analysis by checking the box to fit data to two equations and compare fits (figure 10). The F test to compare fits is only valid when the simpler equation is a special case of the more complicated equation.

Examples of how Prism handles curve fitting can be found in the HMS Beagle Software Solutions columns Analyzing Homologous Competitive Binding Curves with GraphPad Prism, Comparing Dose-Response or Kinetic Curves with GraphPad Prism, and Automatic F Tests to Compare Nonlinear Fits with GraphPad Prism, which offer detailed information on the features of Prism that facilitate data analysis.

Graphing and Presentation Layouts

Prism automatically graphs data sets unless otherwise indicated by the user. The five kinds of graphs available include: XY graphs, bar graphs, and three types of columns graphs: scatter plots, bar plots, or box and whiskers plots. Graphs are fully customizable once generated. Useful plotting features include the ability to generate graphs that have two Y axes and axes that are discontinuous.

Figure 11

Presentation layouts allow up to nine graphs to be placed on a single page in either portrait or landscape orientation. The most notable presentation feature is that graphs, data, and results tables can be combined effortlessly on the same page (figure 11). Slide making can be done either by printing from Prism to a slide-maker, or by using Microsoft PowerPoint. While layouts and graph attributes are maintained when graphs are imported or copied into PowerPoint, it appears that the graphs and layouts may need to be rescaled afterwards. Overall, however, it is obvious from the flexibility available for customizing the graphs' appearance and the features provided for presentation layouts that Prism was designed by scientists for scientists.

Limitations

For users of the Windows version of Prism switching to or sharing data with someone who uses Prism for Macintosh, there are a few differences in presentation and some export and import options that are not available in the Macintosh version. These differences are well documented in an appendix included with Prism, and can be overcome if one is aware of them. One significant difference is that with the Macintosh version, only PICT files can be imported and exported, whereas the Windows version supports the WMF, BMP, PCX, TIF, and GIF file formats. In general, importing and exporting options for data and graphics are fairly restricted for both platforms; dynamic data exchange (DDE) and object linking and embedding (OLE) are not yet supported. Finally, Prism was not designed to handle data-filtering analyses such as fast Fourier transforms (FFT), has no peak-finding capabilities, and cannot perform multiple regression calculations.

Documentation and Online Support

Figure 12

The user's manual is an example of how program manuals should be written. It gives a range of examples of how to set up and perform the various types of statistical analyses included in Prism. For new users, there are many additional aids within the program to help one navigate through analyses, such as New User Hints (figure 12), program opening tips, and a tutorial with sample data. The tutorial takes about 30 minutes and familiarizes new users with the program's main features. Extensive online help is available from the GraphPad Prism Web site and is continuously updated. In addition to Prism program support, the site includes examples of how to analyze specific types of data from homologous binding curves and radioligand binding data and compare dose-response and kinetic curves.

The Bottom Line

Statistical analysis is an intimidating yet necessary task in evaluating data sets. Prism provides a friendly and intuitive interface to help users choose and carry out appropriate statistical analyses. One notable feature is that the program automatically updates graphs and calculations when data sets are modified, and yet also allows users the option of freezing calculations beforehand for comparison. Another bonus is the ability to exclude outlying data points without removing them from the data set or losing prior results and graphs. In addition to its wide array of statistical analyses, Prism goes head to head with many of the common graphing packages for curve fitting and data presentation. Prism's combination of biostatistical analysis, curve fitting, and presentation graphics make it a worthwhile investment. Few programs can match the exceptional implementation of this combination of presentation features and analyses. Prism is a well-rounded statistics package, nicely geared for biomedical research environments.

System Requirements

Prism requires a Macintosh with a 68030 (or better) or PowerPC processor chip running OS 7.1 or later with at least 8 Mb of RAM and 4 Mb of available hard drive space. Both 16- and 32-bit versions are available for Windows 3.1 and Windows 95/Windows NT, respectively. Minimum hardware requirements are a 386 or better processor, at least 4 Mb of RAM, and 3 Mb of available hard drive space. You will also need video with at least 640 x 480 pixel resolution, and it is recommended that it support at least 16 colors or shades of gray with 32-bit QuickDraw. Prism will use a numerical (floating-point) coprocessor to accelerate mathematical calculations if one is installed; however, it does not require one. The Windows and Macintosh versions of GraphPad Prism share the same file format and have identical features while maintaining the characteristic look and feel of the respective operating systems. The interfaces are identical, and users will have no problems switching between versions.

Purchasing Information

GraphPad Prism is available in both Macintosh and Windows versions and may be purchased online from the GraphPad Software Web site. The price for regular users is $495.00; academic users receive a 10% discount and students receive a 25% discount, as do those who purchase two copies at one time. Additional discounts are available for multicopy and network purchases. GraphPad's software comes with a full 90-day money-back guarantee. GraphPad Software may also be contacted by mail at 5755 Oberlin Drive #110, San Diego, California 92121, by telephone at (800) 388-4723 (U.S.) or (619) 457-3909 (outside U.S.), by fax at (619) 457-8141, or by email via the sales, support, or orders departments. Demo versions of GraphPad Prism for both platforms are available for downloading from the company's Web site. The demos are fully functional except for saving, printing, and exporting data.

Since this review was written, GraphPad has released version 3 of Prism for Windows 95, 98 and NT (but not for the Mac). Visit GraphPad's Web site to read about the new features, and to download a demo version of Prism 3 and its manuals.

Ellen M. Quardokus is a research associate in the Department of Biology at Indiana University, in the laboratory of Yves V. Brun.