Multiple Comparisons of Gene Expressions from Microarray Data
by
Jason C. Hsu
The Ohio State University
Coauthors: Jane Y. Chang and Tao Wang

The analysis of microarray data is typically the first stage of distinct decision-making processes. The process might be the fabrication of a diagnostic or prognostic chip, or the elimination of patient subpopulation prone to adverse events, or mining transcription factors that co-regulate genes involved in a disease process. In this talk, we describe how to couple the analysis of gene expression data with the intended purpose of the microarray experiment.

Toward that end, we discuss the pros and cons of the testing for presence of effects approach with the bounding the magnitude of effects approach. After reviewing concepts of multiple comparison error rates, we use mining for transcription factor binding sequences to illustrate how control of the statistical error rate translates into control of the rate of incorrect biological decision. Finally, we show stepdown testing, a popular form of gene expression analysis, is a shortcut to closed/partition testing. We give a set of conditions in the gene expression analysis setting for such a shortcut to be valid, and indicate that subtleties of such conditions do not seem to have been fully appreciated.

Date received: August 29, 2003