Biomedical Engineering Reference
In-Depth Information
(biological) replicates. For example, the experimental noise of DIGE is extremely low owing
to the internal standard experimental design, enabling statistically powered experiments
with very few biological replicates [23], whereas LC/MS experiments have a relatively
high degree of analytical variation with little to no internal standard methodologies being
employed, often resulting in underpowered experiments unless sufficient biological and
technical replicates are employed [40, 41].
Pooling independent (biological) replicates either to produce sufficient material or to
minimize costs (the 'economics of proteomics') should be done with extreme caution with
respect to the statistical power of the resulting data. Pooling samples can be effective if
the technical variation between samples is very low (e.g. sample preparation, analytical
platforms with low noise) and disastrous if it is high (e.g. biological variation unrelated to
experiment, analytical platforms with high noise). Even with sufficient technical replicates
of the same samples to account for high technical variability of the sample preparation and
analysis, performing the
N
1 experiment on pooled samples assumes that the averaging
of populations is reflective of biological signal and that the technical noise is low. Technical
replicates provide confidence in the result from the tested samples, but do not provide any
confidence to the biological relevance. In some cases it may be valid to create sub-pools
from a larger experiment, but in these cases it is still essential to maintain some degree of
individualization of samples to retain statistical power [68].
=
13.3.4 Conclusions
Each of the three major technology platforms described here provides complementary
strengths to a proteomics experiment. Although it is difficult to hold any one of these
strengths paramount, perhaps most often the decision can come down to what you can see
and how well you can see it. This trade-off between sensitivity (depth of coverage) and
statistical power (is it biologically significant?) is quickly becoming a major focus of inves-
tigation in quantitative proteomics studies (e.g. see Refs [23, 40, 41, 68]). It is often the
case that sample fractionation and enrichment strategies are necessary to provide greater
depth of coverage. However, this increased sensitivity comes at a price, as it introduces
technical variation/noise into the system, thereby lowering the statistical power unless a
sufficient number of technical replicates are analyzed in addition to the requisite number of
biological replicates to provide biological significance to the experiment. Analyzing unfrac-
tionated samples introduces the least amount of technical variation/noise, enabling a much
higher statistical power, but at the cost of overall sensitivity/depth of coverage. Ultimately,
the decision of which approach to utilize is best directed by the nature of the experimental
question being asked, and also by the experimental instrumentation/expertise available. In
the best-case scenarios, a proteomics project will not be limited by the utilization of only
one technology platform.
References
1. Wasinger, V.C., Cordwell, S.J., Cerpa-Poljak, A.
et al
. (1995) Progress with gene-product mapping
of the Mollicutes:
Mycoplasma genitalium
.
Electrophoresis
,
16
(7), 1090-1094.
2. Wolters, D.A., Washburn, M.P. and Yates, J.R. III (2001) An automated multidimensional protein
identification technology for shotgun proteomics.
Analytical Chemistry
,
73
(23), 5683-5690.
Multidimensional protein identification technology (MudPIT).
