Quality Control

1. Visual Microarray Inspection

After scanning, a thorough visual inspection of the entire microarray image should be performed. Possible problems to check for include white speckling, holes, smudges, areas of saturation, high and uneven background, uneven hybridization (dark or saturated areas), and crooked grid alignment on microarrays. If the defects span a large area of the microarray, we recommend running a new GeneChip®. Smaller defects that affect only a few single probes can be manually masked and will not contribute to the calculation of the signal intensity. A useful tool for outlier detection is implemented in the 'dCHIP' program developed by Li and Wong (Li and Wong, 2001). It uses a model fitting procedure to identify array, probe and single outliers. Array outliers refer to an unusual probe response pattern in one microarray, which is different from the probe response pattern seen in the other micorarrays in an experiment. This can be the result of image contamination or saturated signals. Probe outliers are most likely due to cross-hybridization to non-specific genes, whereas a single outlier is usually caused by image spikes affecting only one probe in one microarray. The dCHIP program is free for academic users (www.dCHIP.org). In the past, Affymetrix has replaced defective GeneChips® when the original GeneChip® is sent back to them, together with a picture of the scanned image.

2. Settings

All arrays should be set to the same target intensity (TGT) for comparable evaluation. The current Affymetrix default recommendation is 500 for new scanner settings.

3. Affymetrix GeneChip® internal controls and external spiked-in controls

GeneChips® contain several probes for internal control genes, such as beta-actin and GAPDH, and pre-labeled spiked-in controls. All Eukaryotic Expression Arrays contain probe sets for several prokaryotic genes that serve as hybridization controls. The genes represented are BioB, BioC and BioD from biotin synthesis of E. coli, and Cre from P1 bacteriophage. The four controls are pre-mixed in staggered concentrations and added directly in the hybridization cocktail.
These controls are represented by probe sets that correspond to the 5’ end, the middle region, and the 3’ end of the gene. Comparison of the signal intensities of these probe sets can be used to estimate the integrity and the 3’ bias of the labeled sample. High-quality sample preparations yield a 1:2 ratio of signal intensities between the 5’ and 3’ probe sets. As the ratio increases, less 5’ information is detected.

The external eukaryotic hybridization spiked-in controls (BioB, BioC, CreX) should have appropriate increasing signal intensities from BioB to CreX. Failure to observe this increase may be due to inadequate preparation of the controls before spiking.

4. Background noise and scaling factor

Noise (Q) results from small variations in the digitized signal observed by the scanner as it samples the array's surface. The level of noise is calculated by the software by examining the pixel to pixel variations in signal intensities. The measure of background noise (RawQ) should remain consistent across the experiment and within ±3 points of the median.

The scaling factor (SF) is used to multiply the output of any experiment in order to make its average intensity equal to an arbitrary target intensity (TGT) that is set by the user (see section 2 above, “Settings”). Ideally, the SF would be 1, indicating that the arrays were already at the given TGT. Practically and more importantly, the SF should remain consistent across a given set of arrays at the same TGT. The SF for a given set of arrays should be within a 2-3 fold range. Data with large SFs may still be analyzed but may be indicative of a variety of underlying problems (poor sample quality, poor hybridization, unusual sample RNA population, etc.). Background information (BGD) and standard deviation (SD) may be used to correlate SF information. High standard deviations typically correlate with high SFs.

5. Diagnostic plots

Plotting the *.cel intensities in histogram or box plots is helpful in order to visualize different intensity distributions and saturation, which is seen as an additional peak at the highest log intensity in the plot. However, saturation has mainly been seen at a PMT setting of 100% and is not common at the current PMT setting of 10%.

Reference

Li, C. & Wong, W. H. (2001). Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci USA 98, 31-36