Biology Reference
In-Depth Information
Sources of technical noise in DIGE include gel-to-gel variation
as well as sample handling and labeling variation; however, the use
of the Cy2-labeled internal standard methodology effectively nor-
malizes this variation (
1-5
). Additional sources of technical noise
can be related to sample preparation (e.g., subcellular fraction-
ation), laboratory conditions (e.g., medium composition, ambient
temperature, incubation conditions), and sample procurement
(e.g., tissue dissection, protein extraction). Testing independent
biological replicate samples from each condition is necessary to
distinguish between these different sources of variation, and PCA
is an effective way to visualize the experimental variation on a
global scale and enable the assessment of the major sources of vari-
ation with respect to the experimental conditions.
3.2. Principal
Component Analysis
PCA is a commonly used statistical tool that is capable of reducing
the complexity of multivariable space of an experimental dataset
into the major sources of variation, the principal components.
In the case of DIGE, the variation arises from the expression values
of each resolved protein feature, and these features are registered
across all samples and gels (i.e., the spot patterns are matched and
thereby so are the expression values across the dataset).
One need not to understand all of the underlying mathematics
of PCA, which involve covariance matrices and their resulting
eigenvectors and eigenvalues, to be able to utilize the strong diag-
nostic and discovery capabilities that this statistical test provides.
But an understanding of what PCA is doing to the dataset is neces-
sary to properly interpret the results. Considering a plot of all of
the variables in a multidimensional space (a typical DIGE analysis
could contain close to 1,000 defi ned features each of which is
represented in multiple samples), PCA distills out the major sources
of variation as defi ned by the eigenvectors and eigenvalues associ-
ated with each square covariance matrix calculated. The longest
axis through this cloud of multivariable values is defi ned by an
eigenvector (and associated eigenvalue) as the fi rst principal com-
ponent, with the second principal component by defi nition being
orthogonal (perpendicular eigenvector) to the fi rst. Additional
principal components (each accounting for smaller amounts of
variation) can also be defi ned in complex datasets.
Several software tools are available for DIGE analysis, and two of
the more commonly utilized tools found in published experiments,
DeCyder (GE Healthcare) and Progenesis SameSpots (Nonlinear
Dynamics), shall be used as examples here. Both enable the straight-
forward assembly of complex DIGE experimental designs using
the Cy2-labeled internal standard and independent replicates from
multiple experimental conditions, and both enable the visualization
of global variation using PCA and other multivariate statistics.
Where they differ the most is in the approach to feature detection
3.3. Applying PCA
to DIGE Datasets
Search WWH ::
Custom Search