Assessing Signal-to-Noise in Quantitative Proteomics: Multivariate Statistical Analysis in DIGE Experiments - Difference Gel Electrophoresis (DIGE): Methods and Protocols - page 33

consistent with the anticipated biological differences between samples
as well as identify sample outliers, fouled samples, and even potentially
poor experimental design.
Multivariate analyses such as principal component analysis
(PCA) and hierarchical clustering are most commonly used in this
realm of quantitative proteomics, and both are easily accessible
using commonly employed software suites designed for DIGE anal-
ysis, such as DeCyder (GE Healthcare) and Progenesis SameSpots
(Nonlinear Dynamics) (see Note 4). These multivariate analyses
work essentially by comparing the expression patterns of all (or a
subset of) proteins across all samples, using the variation of expres-
sion patterns to group or cluster individual samples. At the very
least, unsupervised clustering of related samples adds additional
confi dence that a “list of proteins” changing in a DIGE experiment
are not arising stochastically. At the very most, these tools offer
additional insights to the underlying structure of the variation
within a complex multivariable experiment and enable the discovery
of protein expression changes by patterns and within subgroups
that are beyond the scope of simple pairwise tests.
Using PCA as an example, this chapter will cover the steps
taken in typical DIGE analyses to perform multivariate statistical
tests. It will use as examples experiments with varying levels of
both biological signal and technical noise to illustrate measures
taken to evaluate the experimental variation and to derive mean-
ingful information for typical DIGE analyses.

2. Materials

DeCyder software (GE Healthcare, Uppsala, Sweden) and SameSpots
software (Nonlinear Dynamics, Newcastle-upon-Tyne, UK)—the
versions used for the examples were 6.5 and 4.0.3779.13732, respec-
tively—and a suitable computer for running these software packages.

3. Methods

As stated in the Introduction, ultimately what is being measured in
quantitative proteomics experiments is variation . In the best cases,
the measured variation is produced by the biology being manip-
ulated, and insight can be drawn from an understanding of the
proteins and their modifi ed forms that give rise to this variation.
This is the signal . However, there is also signifi cant noise that can
contribute to this variation, and this can result both from the tech-
nology employed to measure the variation as well as from the normal
biological variation that exists between samples but is unrelated to
the biology being manipulated.

3.1. Sources
of Variation