Research Article

Exploiting Identifiability and Intergene Correlation for Improved Detection of Differential Expression

Algorithm 1

Steps in the gene re-ranking procedure.
Input: labeled gene expression matrix
       Desired size of differential gene list
Steps:
Calculate two-sample (unpaired) -statistics as in equation .
Re-index genes such that .
Create vector with elements ( scores) in order of ascending magnitude.
Set (default value ). Set first elements of   to zero ( partition).
Convert to by subtracting each gene's average response within each treatment group.
Compute sample correlation matrix of     (Section 4.4).
Find   as in (21). (See discussion of product   in  Section 5.2.)
Create a list of re-ranked genes in the descending order of the values , in which   is the th
     element of   .
Report genes with the largest re-ordered scores as statistical discoveries.
Output:  List of   (experimentally-determined) most differentially-expressed genes following
      the decorrelation processing.
For convenience, it is assumed that is chosen so that is an integer.
The original vector, , of raw -scores (Step ) is ordered by ascending values of   . This ordering
simplifies the definitions of certain quantities in the formal developments. Note: however, that the
output list (Steps and ) is created according to descending   values. This is a more natural
ordering for the end result as we are interested in only the largest values.