Biology Reference
In-Depth Information
Chapter 4
Assessing Signal-to-Noise in Quantitative Proteomics:
Multivariate Statistical Analysis in DIGE Experiments
David B. Friedman
Abstract
All quantitative proteomics experiments measure variation between samples. When performing large-scale
experiments that involve multiple conditions or treatments, the experimental design should include the appro-
priate number of individual biological replicates from each condition to enable the distinction between a
relevant biological signal from technical noise. Multivariate statistical analyses, such as principal component
analysis (PCA), provide a global perspective on experimental variation, thereby enabling the assessment of
whether the variation describes the expected biological signal or the unanticipated technical/biological
noise inherent in the system. Examples will be shown from high-resolution multivariable DIGE experiments
where PCA was instrumental in demonstrating biologically signifi cant variation as well as sample outliers,
fouled samples, and overriding technical variation that would not be readily observed using standard uni-
variate tests.
Key words:
DIGE, Principal component analysis, Multivariate statistics, Variation, Technical replicates,
Biological replicates
1. Introduction
The complex methodologies used in many quantitative proteomics
studies involving multiple experimental conditions often are com-
prised of small sample sets and underpowered experiments. Ultimately
what is being measured in these experiments is
variation
. Utilizing
multiple, independently derived (biological) replicate samples is the
only way to determine if an observed change is due to variation in
the signal (the biology) or the noise (technical/analytical variation
or normal biological variation that is not associated with the experi-
ment). Technical replicates (repeat analyses on the same samples) are
necessary to control for analytical variation, but when this technical
Search WWH ::
Custom Search