Identifying Possibly Positive Genes in Complex Microarray Studies

• Activity Number: 373
• Type: Topic Contributed
• Date/Time: Thursday, August 15, 2002 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing*
• Author(s): Charles Kooperberg*+
• Affiliation(s): Fred Hutchinson Cancer Research Center

Abstract

Many microarray studies differ in several important ways from traditional case-control type studies, where two groups are compared and the interest is in finding genes that are differentially expressed between these two groups. A cell-line experiment related to Huntington's disease is a good example. For this experiment we have gene chips for several different cell-lines that are grown under different conditions for different lengths of time. The investigators want to identify genes that relate to qualitative hypotheses such as ``the expression level of a gene changes in a particular way in some, but not all, cell-lines.'' In addition, rather than identifying a few genes that express significantly (maybe, Bonferroni corrected), they are interested in a somewhat longer list of genes of which we can be confident contains a fair number of ``real positive genes"-- i.e., for which follow-up experiments with northern blots will confirm the results. I will discuss ways in which we can construct test-statistics for such situations, and how we can evaluate how good such a test-statistic is in identifying real, rather than false, positives.