Premium
A cautionary note on the rank product statistic
Author(s) -
Koziol James A.
Publication year - 2016
Publication title -
febs letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.593
H-Index - 257
eISSN - 1873-3468
pISSN - 0014-5793
DOI - 10.1002/1873-3468.12194
Subject(s) - statistic , rank (graph theory) , product (mathematics) , popularity , statistics , differential (mechanical device) , mathematics , property (philosophy) , combinatorics , psychology , physics , social psychology , philosophy , geometry , epistemology , thermodynamics
The rank product method introduced by Breitling R et al . [2004, FEBS Letters 573, 83–92] has rapidly generated popularity in practical settings, in particular, detecting differential expression of genes in microarray experiments. The purpose of this note is to point out a particular property of the rank product method, namely, its differential sensitivity to over‐ and underexpression. It turns out that overexpression is less likely to be detected than underexpression with the rank product statistic. We have conducted both empirical and exact power studies that demonstrate this phenomenon, and summarize these findings in this note.