Assessing Signal-to-Noise in Quantitative Proteomics: Multivariate Statistical Analysis in DIGE Experiments - Difference Gel Electrophoresis (DIGE): Methods and Protocols - page 44

4. These programs differ mostly in the algorithms used to detect
protein features (boundaries) and gel-to-gel alignment/registration
(e.g., vector-based, image warping). In general, they all provide
powerful analytical tools coupled with univariate (Student's
t -test, ANOVA) and multivariate statistical analyses (e.g., PCA
and hierarchical clustering) that can be extremely benefi cial in
evaluating abundance changes of individual protein features as
well as global expression patterns that can help discern changes
that describe the biological phenotype from those that arise
from unanticipated variation in the experiment.
5. SameSpots employs a strategy of image prealignment followed
by the same spot boundaries applied to all features across a data-
set. Whereas this leads to no missing values in the multivariable
space, it can also condense information from multiple features
into the same spot boundary in regions of feature crowding.
The DeCyder approach defi nes spot boundaries on the different
gels independently, and then employs a matching algorithm to
register features across gels. While this yields a higher resolution
for features in crowded regions, it typically involves more man-
ual feature editing and can also lead to missing values in the
dataset which must then be imputed during multivariable statis-
tics (although there is the option to analyze only data that have
been registered in every gel). Both platforms enable the editing
of spot boundaries and feature registration across the gel set to
abrogate these issues, and ultimately both are capable of deliver-
ing high-caliber quantitative DIGE results using both univariate
and multivariate statistical analyses.
6. It is common to allow for mismatches in the dataset with
DeCyder, thereby including some features that may not have a
proper registration in every gel. However, this leads to missing
values that then must be imputed into the multivariate analysis,
which then imposes assumptions about data that do not exist.
Focusing on 100% matches in DeCyder alleviates this issue at
the expense of loosing real information from an uncorrected
mismatch. Another approach is to construct the base set allow-
ing for one or even two mismatches and then apply a 100%
match fi lter to the data in step 9 in Subheading 3.3.1 .
7. Up to fi ve principal components can be selected in DeCyder,
and these can be displayed in any 2- or 3-way combinations.
Three-way comparisons are displayed on a three-dimensional
projection that can be rotated with the computer mouse
(right clicking and dragging).
8. Perform an ANOVA or Student's t -test in the unfi ltered data-
set along with the PCA and other calculations. Then, select
“fi lter dataset” using this criterion with a p -value set to some
meaningful threshold, such as p <0.05 or 0.01. Then, repeat
the PCA calculation with this new fi ltered dataset selected.