Biomedical Engineering Reference
In-Depth Information
Imperfect Data
As in every other physical system, the data generated by a microarray experiment are imperfect.
Determining the magnitude and pervasiveness of these imperfections is one reason for employing
statistical techniques. One way to conceptualize these imperfections is as noise in the
communications channel. This noise is due to limitations of the equipment, reagents, tissue samples,
and deficiencies in the overall process. Some of this noise is unavoidable, and can at best only be
reduced. For example, microarrays are commonly created on glass slides. However, the glass, like
the coating that allows DNA to adhere to the slides, fluoresces slightly when it is excited by the laser
light used to read spots on the microarray.
Similarly, the background noise level is directly proportional to the ambient temperature, in that all
conductors operated above absolute zero produce thermal or Johnson noise. It's possible to operate
the image sensors and amplifiers associated with reading fluorescence signals from microarrays close
to absolute zero, and thereby significantly reduce the noise level contributed by the electronics
equipment associated with the experiment. However, for most bioinformatics applications this
approach to noise reduction isn't practical.
Variations in the preparation of a microarray can make the accuracy of results questionable. For
example, in preparing a glass slide for spotting, slight variations in the volume of substrate deposited
on the slide, or variations in the chemistry of the substrate, can severely compromise subsequent
analysis. Although some applications of microarray expression data, like genetic mapping, are
associated with binary measurements (either present or absent), most applications benefit from
consistent volumes of materials deposited precisely on the microarray so that at least rudimentary
qualitative measurements can be made.
Sources of variability in microarray spot analysis include the stability of the spotting technology used
to create the microarray and the stability of the environmental conditions. For example, the
reproducibility and accuracy of the robotic assembly that determines the location and volume of DNA
material deposited at each spot are critical factors. Furthermore, the environment, including
humidity, temperature, and amount of particulate matter in the air, can add additional variables that
must be considered. For example, if the relative humidity is too high, then the samples in the
microarray may not evaporate as fast as expected. Because of unavoidable variability in the spotting
process, active areas on microarrays are commonly printed in triplicate to provide an internal control.
Variability in microarray experimental results is also a function of the methods used in the data
acquisition phase of a microarray experiment. For example, the two most popular methods of
capturing data from a microarray are scanning and spotting. In scanning, a laser illuminates each
point in the microarray separately. Variability in the data is commonly due to inaccuracies in
positioning the laser over each area where a spot is expected, as illustrated in
Figure 6-8
. In
addition, there is a tradeoff between the diameter of the excitatory laser beam and the relevance of
the fluorescence data. A beam that is only slightly larger than the expected spot size (high
specificity) theoretically provides the least amount of extraneous fluorescence noise, assuming that
the spot in the microarray is in the expected location, with the reading laser superimposed over the
spot. A wider excitatory beam will control for variability in spot location, at a cost of more chances of
fluoresce from contamination, slide coating, and the underlying glass contributing to the fluorescence
signal.
Figure 6-8. Sources of Variability in Reading Microarray Spots Through
Spotting. The ideal situation (A) is when the excitatory laser beam is tightly
focused on a single microarray spot. However, achieving this level of
perfection requires accurate positioning of both the spot and the reading
equipment. If beam position is off the mark (B), gene expression data will
be underrepresented. Using a larger beam than absolutely necessary (C)
incorporates a full spot in the analysis, even if the spot placement isn't
