Data Acquisition

Scanning and the conversion of resulting images (Fig. 3) into numerical data (''quantification'') is a source of technological data variability. Depending on the system used, scanning any array with inappropriate parameter values will either cause signal saturation or miss out low expressed probes. It is therefore better to scan each array more than once, with different parameter values, and to

Antibody immobilisation

Step 2

Analyte addition

Step 2

Step 3

Addition of secondary (reporter) antibody with fluorescent tag. Followed by fluorescent readout.

Step 3

Fig. 2 Deposition and detection of a single probe on a protein array. The general concept of probe selection, probe deposition, target labeling, binding, and scanning are comparable to DNA microarrays, the main differences lying in the specific method of depositing and labeling proteins. In this case, the target sample or analyte is incubated onto the array before labeling the target by means of a specific antibody with a fluorescent tag. The analyte could also be labeled before incubation; this depends on user preference and the biological system under investigation. (View this art in color at www.dekker.com.)

select a best scan or use statistical approaches to combine the information from multiple scans. A standard protocol must be associated with this approach. This needs to cover suitable scanning parameters, a consistent quantification method, recommended level of manual fine-tuning for the grid alignment, and quality control steps such as the assessment of background noise.

Data Processing

It is most likely that despite use of stringent laboratory and data acquisition protocols, there are still systematic sources of variation or noise in the data that cannot be attributed to the biological sample differences under investigation. The use of data processing techniques[4-6] is not optional but required. Image data is associated with background noise (specks of dust, uneven hybridizations, etc.), which needs to be adjusted. Most array data are assumed to be on a scale that benefits from prior transformation of the data values. The term ''normalization'' in relation to microarray data is best defined as a collection of mathematical methods that are used to make target samples comparable by removing systematic effects introduced by dye-label incorporation differences, hybridization differences resulting in overall image ''brightness'' differences, and differences across an individual array caused by deposition of probes by different robotic print tips.

Background measured by the image processing software is usually removed by subtracting individual probe background from the probe signal value or by subtracting a local average background from each individual probe signal value. It is also recommended to define a detection-threshold limit for each array, which is the mean plus two standard deviations of negative control probes on the array, or another robust estimate for theoretically nonex-pressed probes on the array.

On-chip signal ratios are most often log-transformed to base 2, which symmetrizes the distribution. To stabilize variance, transformations such as arc sinh are still being evaluated. For dual-target arrays, the currently most common method of normalization is the locally weighted scatterplot smoother (LoWeSS), which adjusts for the nonlinear differences between two dye labels on an array.[5] The effect of this can be seen in Fig. 4.

In many cases, the normalized log ratio is still subject to different levels of variance, and to make them comparable across a number of arrays, a second step can be included in the normalization, which adjusts variance rather than location of log ratios. One such method is the median absolute deviation scaling.[6]

Single-target arrays do not normally require any on-chip normalization like the above. The center (median) of the absolute signal intensities on the array is often scaled to match other arrays. Other common forms of normalization take into account specific print tip groups; that is, standard normalization methods are performed on physical subsets of the array rather than on the array as a whole. For further analysis, in particular cluster analysis, normalization also occurs on a per-gene level, which means that each gene's values across a number of arrays are adjusted to have a mean of 0 and a standard deviation of 1.

A special case of normalization is one that estimates the size of an effect rather than correct for it. Given a high enough level of replication, analysis of variance (ANOVA) can provide estimates for each effect (hybridization, dye label, print tip group, etc.) that has been included in the model; that is, the observed signal value is described by the level of signal that is due to unwanted sources of data variation and the signal differences that are due to real biological effects.[7]

Data Analysis

Microarrays produce an amount of data that has rarely been seen before in biological research. It is common to

Getting Started With Dumbbells

Getting Started With Dumbbells

The use of dumbbells gives you a much more comprehensive strengthening effect because the workout engages your stabilizer muscles, in addition to the muscle you may be pin-pointing. Without all of the belts and artificial stabilizers of a machine, you also engage your core muscles, which are your body's natural stabilizers.

Get My Free Ebook


Post a comment