books and software
Design, Analyze, and Optimize with DesignExpert
mation provided in the displays. DesignExpert organizes these tasks in subsections of a folder for each project and set of experimental data. Most DoE operations are intuitive and seamless, supported by useful information in graphics and text. With minimum effort—and often minimal computational time (PC with 350-MHz Pentium II microprocessor, 64-Mb RAM)—virtually any part of a DoE project can be examined and modified, and the results can be evaluated. All text and graphics can be easily cut and pasted into reports, edited, and rearranged. In general, this option is much more powerful than the ability to export text.
factors. A large number of factors would be desirable for DoE applications in combinatorial chemistry. A large number of responses would be useful for DoE apDesign-Expert6.0.6 plications where there are many target Stat-Ease, Inc. analytes but relatively few instrumental $995, network licenses available on a conditions to optimize in high-resoluconcurrent usage basis tion chromatography or spectroscopy, www.statease.com for example. Technically, several factorial Requires: Windows 95 and up designs are restricted to categorical variables, but these often can be reclassified esign-Expert 6.0.6 for design-ofas numeric variables without hindering experiment (DoE) can serve a the computations. variety of needs among experimental In the software’s analysis section, the chemists, from screening to modeling goal is to reduce the mathematical model and optimization. This review highlights to key factors controlling observed rethe computational capabilities and funcsponses. Throughout this section, results tions of Design-Expert 6.0.6 and some of statistical analyses are of the less obvious experiA: Water A: Water automatically displayed in mental conditions. Copies 5.0 5.0 2 2 text and graphic formats of Design-Expert can be to provide consistent, obdownloaded from www. jective criteria for fitting statease.com for a 30-day the model to the experitrial. The software is well 4 mental results. Frequent supported by online help 4 2.0 2.0 2.0 2.0 40 annotations are provided features and a printed 3.5 3.5 50 50 2 2 to flag statistically signifimanual, which includes 17 Design points 60 60 70 cant results that have beartutorials and 5 appendices 2.75 70 3 80 80 ing on subsequent choices of technical information 3.5 4 2 2 2 2 for the model. As might for advanced DoE applica4.0 3.0 4.0 4.0 3.0 4.0 be expected, the computations. Mathematical methB: Alcohol C: Urea B: Alcohol C: Urea Viscosity Standard error of viscosity tional time for the analysis ods for DoE analysis are section starts to become Contour plots from a mixture design for three factors (water, 3–5%; alcohol, the focus of recent books noticeable as the complexi2–4%; urea, 2–4%; constrained to a total of 9%) and two responses (viscosity, by D. Montgomery (1) ty of the problem increases. 35- to 144-mPa-s; turbidity, 323–1122 ntu): left, the dependence of the “visand C. F. Wu and M. A notable feature of the cosity” response; and right, the standard error of the “viscosity” response, Hamada (2). Users can optimization section is conas functions of the solvent composition. The 14 design points are clearly loget the educational vervenience. Responses can be cated on axes from each vertex to the midpoint of the opposite side. The sion, Design-Ease, on CDcalculated for any combinaaxes represent the range of values for each factor: two center points, two ROM with a textbook by tion of allowed experimenM. Anderson and P. Whit- points at each of the three vertices, three center edge points, and three tal conditions. The user comb (3). The educationaxial check blend points. With very little modification, this design could easalso has a choice of graphic al version of Design-Expert ily represent the optimization of a ternary solvent mobile-phase mixture for or numeric displays for opcomes with Montgoman HPLC separation. timization. Procedures to ery’s textbook. set desired values or criteDoE methods use probria (maximum, minimum, in range) for New designs are selected from four ability and statistics to define the minithe factors and responses are self-explanacategories: factorial, response surface, mum number of experiments needed to tory. A table lists values of the factors and mixture, and crossed-process mixture. identify significant cause-and-effect relaresponses for all acceptable solutions, and Page layouts are well organized and intionships between a given number of factors and one or more responses. Math- formative. Designs are displayed on sep- the results can be copied into reports. arate pages in a table format. DesignResults for each solution can be examematically, DoE methods are well develExpert limits many designs to a relatively ined in a graph one at a time, but it is oped to identify efficient experimental small maximum number of factors, but trivial to step through all possible soludesigns. The three major components all designs can have up to 100 responses. tions and see the differences. Because a of DoE are design, analysis, and optiIn contrast, the DoE module in Statistilarge quantity of information on optimimization. The strength of Design-Expert ca by StatSoft, Inc. (www.statsoft.com) zation can be produced in a very short is the ease with which these tasks can be allows designs to have more than 100 time—for various conditions of interest— carried out and the usefulness of infor-
D
222 A
A N A LY T I C A L C H E M I S T R Y / A P R I L 1 , 2 0 0 2
books and software
capabilities to visualize the results are very important. In retrospect, several DoE conventions may be nontrivial to implement experimentally, depending on the user’s circumstances. For example, DesignExpert follows statistical conventions to avoid bias and assigns a random order in which to conduct experiments. Moreover, DoE experimental conditions are defined to create a geometrically uniform distribution over the range of values specified by the user, as shown by the example on p 222 A. Response surface and mixture designs in Design-Expert often use five or six levels per factor: Assigned values are not just whole integers but also include one-third and two-third fractions, in or beyond a specified range, to maintain geometric symmetry. Depending on the design, suggested values in a range from 2 to 4 may be 2, 2.33, 3, 3.33, 3.67, 4 or 1.66, 2, 3, 4, 4.34. These conditions are a clear departure from the common experimental sequences, which vary logarithmically or by orders of magnitude. It is expected that users will follow a learning curve in the application of DoE methods to their experiments. The most basic decisions are which design to use and how many levels to set for the known factors. If some experimentalists eventually find themselves searching for capabilities beyond Design-Expert, they will most likely still value the computational experiences that increased their understanding.
Reviewed by Katherine T. Alben, New York State Department of Health, Albany, NY.
Books and Software Received b c b Engineering Biosensors:Kinetics and Design Applications Ajit Sadana Academic Press 2002, 404 pp, $89.95
points. Tecplot 9.0 is designed to render 3-D data up to 100⫻ faster than version 8.0, and its new true color and translucency make for more impressive plots, animations, and presentations.
Engineering Biosensors explores the quantitative process of receptor–analyte binding using fractals, fractal kinetics, and fractal dimensions. Numerous examples of the mathematical techniques, most notably Havlin’s equation, are given for both the association and the disassociation between analytes and receptors. Later chapters explore DNA hybridization reactions using biosensors and surface plasmon resonance biosensors.
c
b M etalNanoparticles:Synthesis, Characterization,and Applications Edited by Daniel L. Feldheim and Colby A. Foss, Jr. Marcel Dekker 2002, 338 pp, $150 This book discusses the optical, electronic, and structural properties of nanoparticles, particularly how their size, shape, and surface chemistry affect their performance. Because the nature of these properties has only just come to be understood, this text seeks to unify the themes of metal nanoparticle research.
c Tecplot9.0 Amtec Engineering, Inc. www.amtec.com Requires: Windows 95, 98, 2000, NT, Linux, and UNIX $1395 (base), academic discounts available
References (1)
(2)
(3)
Montgomery, D. Design & Analysis of Experiments, 5th ed.; John Wiley: New York, 2001. Wu, C. F.; Hamada, M. Experiments: Planning, Analysis, and Parameter Design Optimization; John Wiley: New York, 2000. Anderson, M.; Whitcomb, P. DoE Simplified: Practical Tools for Effective Experimentation; Productivity, Inc.: Portland, OR, 2000.
Tecplot 9.0 is the technical plotting software package that takes advantage of affordable graphics cards available on today’s PCs and UNIX workstations. OpenGL has been incorporated into the software, which can be used to interactively visualize, explore, and quickly analyze large 2-D and 3-D data sets consisting of millions of data
Origin 7.0 OriginLab www.originlab.com Requires: Windows 95, 98, 200, NT4 $699 Origin 7.0 combines presentation-quality graphics, C language, and other elements. Interface improvements include modernized text and drawing tools, Plotting Wizard with Graph Gallery, and Nonlinear-Curve Fitting Wizard. Advanced analysis features include twoway ANOVA, survival analysis, and nonlinear-curve fitting with automatic parameter initialization. A new Code Builder interface is added to the software to provide a C programming and debugging environment.
c CyberLAB Know ledge Engineering System 2.0 Scientific Software, Inc. www.scisw.com Requires: SQL Server 2000, Oracle 8i, and Windows 2000 with IIS 5.0; clients need Internet Explorer 5.01 or higher, Windows 95, 98, 2000, NT, ME $1500–6500 each (based on concurrent users) This Web-based version of CyberLAB Knowledge Engineering System enhances compliance with the U.S. Food and Drug Administration’s stipulations for electronic signature and electronic submission by meeting the requirement of 21 Code of Federal Regulations, Part 11. Features include a new user interface design that tailors file display pages. It offers unlimited userdefined keys for electronic records and supports Apple Macintosh files, SQL Server 2000, and Oracle 8i. The software also supports clustered Web servers to ensure maximum uptime and stores files in a single logical unit.
A P R I L 1 , 2 0 0 2 / A N A LY T I C A L C H E M I S T R Y
223 A
