|
Abstract:
|
Use of microarrays for pharmacogenomics should meet accepted statistical standards such as sensitivity and specificity requirements. We will first give recommendations for statistically designing microarrays and sample hybridization toward reproducible results, designs that have been implemented in a collaborative cancer recurrence prognosis project. Then, multiple testings ANALYSIS issues of gene expression levels will also be discussed. These issues include the definition of Type I error rate (for a single hypothesis), control of the False Discovery Rate (FDR) versus generalized Familywise Error Rate (gFWER), conditions for shortcutting closed/partition tests to step-down and step-up methods, and a subtle condition required for the validity of permutation tests.
|
