Biomedical Engineering Reference
In-Depth Information
Basics
The overview of a typical microarray experiment underscores the dependence of bioinformatics work
on an awareness of error sources and variability so that statistical methods can be used to control for
their effects on experimental results.
Randomness
One of the key statistical concepts highlighted by the microarray experiment is that data are
inherently noisy and that randomness is inherent in any sampling process. Furthermore, randomness
is inherent in, and a necessary component of, biological systems. Whereas the randomness in
mechanical systems and electronic circuitry is often minimized as much as is economically possible,
randomness is an integral component of the workings of biological systems. Mutations and the
distribution of maternal and paternal genetic material during meiosis are biological processes that
reflect the dependence of biodiversity on the randomness of biological processes.
Every measurement system introduces noise—random variability—into the desired signal. This noise
can be minimized by controlling the external environment (for example, by reducing the ambient
temperature in a system designed to make very low-level measurements), or, more often, by
reducing the bandwidth of the system, using statistical techniques. For example, by reducing the
bandwidth of acceptable (good) data, it can be more readily differentiated from bad data and made
more apparent and available. Even though statistical techniques can be used to filter data during the
final analysis of a gene expression experiment, reliance on statistical analysis of the final results
alone isn't optimal. For example, although analysis of intra-array spot fluorescence intensity can be
used to control for contamination and other sources of variability, a better approach is to minimize
variability in the overall process. As a result, there will be more experimental data, and less need to
run controls that add to the experimental overhead without contributing directly to gene expression
discovery.
The microarray experiment also illustrates how conventional mechanical systems are more variable
than their electronic counterparts. Compared to computers and other so-called finite-state machines
defined in silicon and software, conventional mechanical systems such as robotic arms and micro-
pipettes are much more variable in their operation. One of the greatest potential sources of variability
in the placement of cDNA solution on a prepared glass slide microarray is the robotic assembly that
performs the spotting of the microarray. What's more, the amount of cDNA that actually adheres to
the slide can vary widely as well, as a function of the slide coating, the ambient environmental
conditions, and the presence of contaminants. Estimating the variability contributed by the
mechanical and biochemical systems—through computer modeling or direct measurement—provides
an indication of the expected value of the data. Nanotechnology may eventually reduce the variability
of computer-enabled mechanical systems to the point that it is comparable to that of digital
electronic circuitry.
Variability Is Cumulative
Regardless of whether the source is mechanical, biological, or electronic, variability is cumulative, in
that noise introduced in the early stages of a system propagates and is amplified by later activity in
the system. For example, extraneous genetic material commingled with the cDNA used to create a
microarray will add to the fluoresce activity measured from each spot. This not only adds to the noise
level of the system and decreases the effective dynamic range of the experiment, but the fluoresce
activity at otherwise quiescent locations in the microarray will be amplified by the PMT or CCD-based
system and digitized. Unless the variability can be quantified through control experiments, the gene
expression conclusions suggested by the data analysis will be incorrect.
Controlling variability is a key component of process management. Managing the chain of processes
in the microarray experiment involves controlling variability through computer-enabled statistical
