Statistical Analysis of Gene Expression Microarray Data

Rating:
Summary: Outstanding survey
Review: Microarray studies are becoming the preferred research tools in many areas, including cancer research, development studies, and studies in organisms' responses to their environments. Because of differences between organisms or between experiments, microarray data is always statistical in nature. The problem is that the data aren't well suited to traditional statistics. Instead of studying a few characteristics in large numbers of individuals, microarray studies typically yield thousands of data values for a few dozen samples.

That mismatch, between current statistical practice and microarray analysis requirements, seem to be driving many innovations in statistical analysis. This book is a brief survey of four of those areas of analysis: model-based analysis, experimental design, classification, and clustering.

The first section, on model-based analysis, is brief. Mostly, it seems to establish the language used in later sections. The next, on experimental design, deals with ways for getting the most information out of the fewest samples. The costs of arrays and processing are dropping, but still high. More analysis on less data makes good economic sense. The DNA samples analyzed also have costs - some can only be prepared in minute amounts, others must be extracted surgically from human patients. Either way, it's important to maximize the knowledge harvested from limited amounts of biologcal material.

The next section, on discrimination, is a bit longer. It briefly summarizes a wide variety of techniques for deciding which category best represents any one sample. This section gives a good review of analytic approaches: Fisher classifiers and their descendants, principal components, support vectors, and decision trees. Within trees, the authors note that the number of missing values in typical microarray data may interfere with standard analysis, and that surrogate variables may be needed in many cases. AI and data mining techniques aren't broadly represented, but this chapter is still very informative.

The final section, on clustering, was shorter. It was reasonably informative, and I gleaned a few new facts from it. Mostly, though, it seemed to present techniques that are already well known.

This book is a survey, so it emphasizes breadth over depth. Many algorithms described only briefly, and some are just mentioned by name. The developer will need to chase references to find an implementable level of detail. Still, the book has value as an index to references and as a comparison of techniques.

//wiredweird

Rating:
Summary: Excellent book for data analyst
Review: Thorough converage of statistics involved in microarray data analysis. It presents important knowledge for biologists who use data analysis tools but would like to know what is behind the scene. Understanding the book needs some statistical background and hence not a easy book for biologists and genetists who do not have that knowledge.
I would like to emphasize that experiment design issue is presented in a very clear way and should be read by all who plan to start project related to gene expression. Clustering and classification are two major analysis methods for microarray data, and the comprehensive discussion of the statistical mechanisms for each method in the last two chapters will help analysts to choose the right methods when mining the data. The first chapter seems to be a little out of the place, because it mainly discusses model-based genechip data analysis. This chapter touches a little about preprocessing and gene selection but far from complete.
A chapter with thorough discussion of pre-processing techniques and gene selection techniques would make this a prefect book. Overall it is a great reference for anyone who is interested in microarray data analysis!