Catálogo de publicaciones - libros

An overview of industrial screening experiments is presented, focusing on their applications in process and product design and development. Concepts and terminology that are used in later chapters are introduced and explained. Topics covered include a discussion of the general framework of industrial experimentation, the role in those activities played by screening experiments, and the use of two-level factorial and fractional factorial designs for screening. Aliasing in fractional factorial designs, regular and nonregular designs, design resolution, design projection, and the role of confirmation and follow-up experiments are discussed. A case study is presented on factor screening in a plasma etching process from semiconductor manufacturing, including a discussion of the regular fractional factorial design used and the analysis of the data from the experiment.

Reduction of the variability in performance of products and manufacturing processes is crucial to the achievement of high levels of quality. Designed experiments can play an important role in this effort by identifying factors with , that is, factors that affect performance variability. Methods are presented for the design and analysis of experiments whose goal is the rapid screening of a list of candidate factors to find those with large dispersion effects. Several types of experiments are considered, including “robust design experiments” with noise factors, and both replicated and unreplicated fractional factorial experiments. We conclude that the effective use of noise factors is the most successful way to screen for dispersion effects. Problems are identified that arise in the various analyses proposed for unreplicated factorial experiments. Although these methods can be successful in screening for dispersion effects, they should be used with caution.

Pooling experiments date as far back as 1915 and were initially used in dilution studies for estimating the density of organisms in some medium. These early uses of pooling were necessitated by scientific and technical limitations. Today, pooling experiments are driven by the potential cost savings and precision gains that can result, and they are making a substantial impact on blood screening and drug discovery. A general review of pooling experiments is given here, with additional details and discussion of issues and methods for two important application areas, namely, blood testing and drug discovery. The blood testing application is very old, from 1943, yet is still used today, especially for HIV antibody screening. In contrast, the drug discovery application is relatively new, with early uses occurring in the period from the late 1980s to early 1990s. Statistical methods for this latter application are still actively being investigated and developed through both the pharmaceutical industries and academic research. The ability of pooling to investigate synergism offers exciting prospects for the discovery of combination therapies.

Twenty years ago, drug discovery was a somewhat plodding and scholastic endeavor; those days are gone. The intellectual challenges are greater than ever but the pace has changed. Although there are greater opportunities for therapeutic targets than ever before, the costs and risks are great and the increasingly competitive environment makes the pace of pharmaceutical drug hunting range from exciting to overwhelming. These changes are catalyzed by major changes to drug discovery processes through application of rapid parallel synthesis of large chemical libraries and high-throughput screening. These techniques result in huge volumes of data for use in decision making. Besides the size and complex nature of biological and chemical data sets and the many sources of data “noise”, the needs of business produce many, often conflicting, decision criteria and constraints such as time, cost, and patent caveats. The drive is still to find potent and selective molecules but, in recent years, key aspects of drug discovery are being shifted to earlier in the process. Discovery scientists are now concerned with building molecules that have good stability but also reasonable properties of absorption into the bloodstream, distribution and binding to tissues, metabolism and excretion, low toxicity, and reasonable cost of production. These requirements result in a high-dimensional decision problem with conflicting criteria and limited resources. An overview of the broad range of issues and activities involved in pharmaceutical screening is given along with references for further reading.

Microarrays are an important exploratory tool in many screening experiments. There are multiple objectives for these experiments including the identification of genes that change expression under two or more biological conditions, the discovery of new cellular or molecular functions of genes, and the definition of a molecular profile that characterizes different biological conditions underlying, for example, normal or tumor cells. The technology of microarrays is first described, followed by some simple comparative experiments and some of the statistical techniques that are used for their analysis. A very important question arising in the design of screening experiments with microarrays is the choice of the sample size and we describe two approaches to sample size determination. The first approach is based on the concept of reproducibility, and the second uses a Bayesian decisiontheoretic criterion to make a trade-off between information gain and experiment costs. Finally some of the open problems in the design and analysis of microarray experiments are discussed.

Living organisms need proteins to provide structure, such as skin and bone, and to provide function to the organism through, for example, hormones and enzymes. Genes are to proteins after first being to messenger RNA. Even though every cell of an organism contains the full set of genes for that organism, only a small set of the genes is functional in each cell. The levels at which the different genes are functional in various cell types (their ) can all be screened simultaneously using microarrays. The design of two-channel microarray experiments is discussed and ideas are illustrated through the analysis of data from a designed microarray experiment on gene expression using liver and muscle tissue. The number of genes screened in a microarray experiment can be in the thousands or tens of thousands. So it is important to adjust for the multiplicity of comparisons of gene expression levels because, otherwise, the more genes that are screened, the more likely incorrect statistical inferences are to occur. Different purposes of gene expression experiments may call for different control of multiple comparison error rates. We illustrate how control of the statistical error rate translates into control of the rate of incorrect biological decisions. We discuss the pros and cons of two forms of multiple comparisons inference: testing for significant difference and providing confidence bounds. Two multiple testing principles are described: closed testing and partitioning. Stepdown testing, a popular form of gene expression analysis, is shown to be a shortcut to closed and partitioning testing. We give a set of conditions for such a shortcut to be valid.

The role of projection in screening is discussed and a review of projection properties of factorial designs is provided. The “projection of a factorial design onto a subset of factors” is the subdesign consisting of the given subset of factors (or, equivalently, the subdesign obtained by deleting the complementary set of factors). A factor-screening design with good projections onto small subsets of factors can provide useful information when a small number of active factors have been identified. The emphasis in this chapter is placed on projection properties of nonregular designs with complex aliasing. The use of projection in search designs and uniform designs is also discussed briefly.

Supersaturated designs are fractional factorial designs that have too few runs to allow the estimation of the main effects of all the factors in the experiment. There has been a great deal of interest in the development of these designs for factor screening in recent years. A review of this work is presented, including criteria for design selection, in particular the popular E() criterion, and methods for constructing supersaturated designs, both combinatorial and computational. Various methods, both classical and partially Bayesian, have been suggested for the analysis of data from supersaturated designs and these are critically reviewed and illustrated. Recommendations are made about the use of supersaturated designs in practice and suggestions for future research are given.

The idea of using a group screening procedure to identify the important or active factors using a small designed experiment was described by Watson (1961) and is now applied in a variety of areas of science and engineering. Watson’s work built on the earlier ideas of Dorfman (1943) for screening pooled samples of blood in order to identify diseased individuals using minimal resources. Generalizations and extensions of Watson’s technique have been developed by a number of authors who have relaxed some of the stringent assumptions of the original work to make the methods more widely applicable to real problems. An overview of some of the proposed screening strategies is presented, including the use of several stages of experimentation, the reuse of runs from earlier stages, and screening techniques for detecting important main effects and interactions

The problem of designing an experiment for selecting a good model from a set of models of interest is discussed in the setting where all factors have two levels. The models considered involve main effects and a few two-factor interactions. Two criteria for the selection of designs for model screening are introduced. One criterion selects designs that allow the maximum number of distinct models to be estimated (estimation capacity). The other maximizes the capability of the design to discriminate among competing models (model discrimination). Two-level orthogonal designs for 12, 16, and 20 runs that are optimal with respect to these criteria are constructed and tabulated for practical use. In addition, several approaches are discussed for the construction of nonorthogonal designs. The chapter includes new results on orthogonal designs that are effective for model discrimination.