Study Design in DIGE-Based Biomarker Discovery - Difference Gel Electrophoresis (DIGE): Methods and Protocols - page 203

In case of a change of the analytical platform, all samples of the
discovery experiment should be reanalyzed using the validation
platform. This is especially necessary for immunological methods
to prove the ability of the antibodies to quantify the potential
biomarker.

To validate the potential biomarkers, it is advisable to collect at
least the same number of new samples as in the discovery experi-
ment. Many researchers use the validation experiment to estimate
a threshold value for each biomarker for future studies. An expres-
sion level above or below such a threshold could then be used in
future to allocate an unknown sample to an experimental group
(e.g., disease or control). In this context, a much larger number of
samples should be included in the validation experiment than in
the discovery experiment (4-10 times can be found in literature,
clearly the larger the better).

3.2.2. Sample Size of the
Validation Experiment

The number of potential biomarkers tested in the validation exper-
iment is much smaller than the number of spots investigated in the
discovery set. Nevertheless, the resulting p values of the corre-
sponding t test should be adjusted for multiple testing. Often it is
suggested to apply the control of the FWER rather the FDR (see
Note 9). Those proteins which show also in the validation experi-
ments a signifi cant difference in their expression level between the
experimental groups can be considered as validated biomarkers.

3.2.3. Statistical Evaluation
of the Results of the
Validation Experiment

4. Notes

1. A level of 0.05 is generally accepted in clinical routine
diagnostics. Therefore, it is reasonable to use the same signifi -
cance level in a proteomic discovery experiment. Using a
smaller level (e.g., 0.025 or 0.01) decreases the power of the
statistical test, and several biomarkers might not be found.
2. In a biomarker study, the researcher wants to fi nd one or more
protein spots, whose expressions are associated, for example,
with the incidence of a disease. Due to the type 2 error, some
of the proteins declared as statistically not signifi cant (adjusted
p value of >0.05) might in reality correlate very well with the
disease. However, the needs of the researcher are satisfi ed as
long as a suffi cient number of true-positive proteins have been
found. Thus, fi xing a moderate level b for the type 2 error

(e.g., 0.2 which is also applied in conventionally clinical studies)
in the sample size calculations is acceptable. Note that applying
a larger level (and therefore a lower statistical power) increases
the false-negative rate so far that only a few (or maybe no) true
biomarkers might be found in the analysis.