Study Design in DIGE-Based Biomarker Discovery - Difference Gel Electrophoresis (DIGE): Methods and Protocols - page 197

Table 1
List of statistical terms

Term Symbol Defi nition

In the single-hypothesis setting

Type 1 error a Probability that a protein is identifi ed, which in truth, is not

related to the clinical outcome

Type 2 error b Probability that a protein is not identifi ed, which in truth, is

related to the clinical outcome

Power 1 − b Probability that a protein is identifi ed, which in truth, is related

to the clinical outcome

In the multiple-hypotheses setting
Family-wise type
1 error rate

FWER Probability that at least one of the proteins, which are in truth

not related to the clinical outcome, is identifi ed

False discovery rate FDR Expected proportion of identifi ed proteins, which are in truth not

related to the clinical outcome, among all identifi ed proteins

Power 1 − b Expected proportion of identifi ed proteins, which are in truth,

related to the clinical outcome

Sample size n Sample size per group

Standard deviation s Common standard deviation (assumed to be equal in both groups)

Effect size q Difference in group means times the common standard deviation

m Number of tested hypotheses (proteins)

m 1 Number of effective proteins, in truth, related with the clinical

outcome

p 1 Proportion of proteins, which are in truth related to the clinical

outcome, among all investigated proteins

D Difference of group means of the expression values

proteins. Testing about 20 proteins simultaneously, each with a
0.05 level, we already expect 1 protein to be a false-positive (if, in
truth, all 20 proteins are not related with the clinical outcome). If
about 3,000 proteins are tested and assuming none of them to be,
in truth, related with the clinical outcome, again using a 0.05 level
for each test would lead to 150 false-positive decisions.
To avoid these pitfalls, a number of multiplicity adjustment
procedures are available (see refs. ( 4, 5 )). These adjustment pro-
cedures generally result in so-called adjusted p values (instead of
the standard p values) that are to be compared to the signifi cance
level 0.05. Literature is commonly discussing methods for gene
expression studies; however, for proteomic studies, these methods
can generally be applied. Two approaches are widely used to adjust
for multiplicity:

1. Control of the family-wise error rate (FWER), i.e., the proba-

bility to identify at least one protein that is not related with the
clinical outcome.