Statistics for Microarrays : Design, Analysis and Inference

Rating:
Summary: Very good, on the whole
Review: This is the best introduction I know for anyone trying to learn the bioinformatics of microarrays. It starts with a brief description of the DNA microarrays, their chemistry, and the sources of uncertainty in their measurements, just enough for a non-biologist to get the general ideas. It skips the steps of scanning and spot recognition, mostly, and jumps right into analysis of the array of spot readings.

That is where the text comes into its own. One happy surprise is the book's emphasis on quality control and error management. Quality issues are addressed first by themselves, then as they affect the design and analysis of an experiment's biological meaning. This covers a wide variety of issues, including dye swaps, array background correction, and inference in the presence of low-quality data. There are soft spots in the discussion, especially in handling of missing data. That fits the general tone of the book, though, by stressing understanding over rigor.

This book comes with a macro package for the R environment, an open-source system somewhat like Matlab or Mathematica. That is both the strength and the weakness of this book. The strength of course, is the working code. It lets you see a real implementation of the algorithms that the authors describe. The weakness is that the implementations don't explain how the algorithms were developed, why they work, or how to recognize when they've been pushed past their breaking points. If you need more than rote recitation of the authors' implementation, you may find this frustrating. Also, the book uses five data sets for concrete discussion, but the software kit seems to include only one.

Microarray data sets (a few individual with thousands of measurements each) are very different from standard statistical data sets (lots of individuals with few measurements each). Despite the dramatic improvements of the last few years, the processing of the arrays themselves still varis widely under even the tightest control. Microarrays really do need different kinds of analysis and experimental design. This is a very readable explanation of why and how those procedures are used. I just wish the procedures themselves were presented in a little more depth.

//wiredweird