Controlling False Discovery (FD) When Planning Microarray Experiments

• Activity Number: 334
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Eric R. Siegel*+ and Trey Spencer and Rudolph S. Parrish
• Companies: University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and University of Arkansas For Medical Sciences
• Keywords: microarray ; multiple comparison ; false discovery ; planning

Abstract

Sophisticated resampling methods exist to control FD in microarray experiments once the data are in hand. Less attention has been paid to control of FD during the design stage. In this context, Simon proposes controlling Expected FD for tests on k genes as follows: simply choose the level p of test such that kp =< u, where u is the desired Expected FD. We propose a method almost as simple as Simon's for controlling Actual FD. Since FD is binomially distributed, one can calculate the exact binomial probability Pr{FD =< u|Bin(k,p)} of having u or fewer false discoveries. One can then adjust the level p of test needed to raise this probability to a desired "1 - alpha" confidence level, thus controlling Actual FD. For u=0, our method reduces to the Sidak multiple-comparison adjustment. For u>0, the level p of test can be approximated to twelve decimal places using Newton's method in an Excel spreadsheet; convergence typically takes about six iterations.