g

g

,

s, which may be differential, were missed (hence the Type II

ue to the high extreme outliers occurring in the cancer replicates.

. A high extreme outlier in gene 216987_at in GDS3139 missed a true DEG.

PA

er profile outlier analysis algorithm (COPA) was perhaps the

ne for discovering heterogeneous DEGs [Tomlins, et al., 2005].

as modified the t statistic as a ratio of the distance between the r^th

value was the 9^th) percentile of the case expressions and the

of all expressions over the median absolute distance (deviated

whole population median of the gene). The COPA t statistic is

s below,

ݐ^஼ை௉஺ൌ^{ݍ௥ሺܡሻെߣ}

ߪ

(6.10)

ൌ1.4826 ൈmedianሼܠെߣ, ܡെߣሽ, ܠ stands for a vector of the

xpressions and ܡ stands for a vector of case expressions, ݍ௥ሺܡሻ

r the r^th percentile of ܡ and ߣ is the median of both ܠ and ܡ. The

on of the r^th percentile is irrelevant to the number of outliers.

e outlier number is small, the difference between ݍ௥ሺܡሻ and ߣ is

owever, when the outlier number is large, the difference between

d ߣ is large. The r^th percentile also depends on the distribution of

ta set. The COPA p values are calculated by the permutation

.