Researchers from the University of Essex in Colchester, UK, have provided an overview of image processing methods for Affymetrix GeneChips. They show that correcting for sources of error using image processing methods results in a significant increase in sensitivity.

The authors describe methods for correcting for errors that occur while scanning the chips. They touch upon image segmentation methods for aligning probe cells with pixels, and they discuss errors stemming from pixel saturation, from pixels recording or losing signal from the surrounding region and from the location of pixels relative to the probes. They note that the 75th percentile of pixel intensities in the cel file provides the most accurate representation of gene expression in the raw data stored in the dat file.

The authors describe a package of algorithms called Harshlight, which minimizes irregularities such as specks of unwanted matter and blemishes, as well as a method for removing spatial blemishes based on mismatch probe adjustment. They also note a program called Microarray Blob Remover that corrects for blobs that appear in the new generation of single nucleotide polymorphism, exon and tiling arrays. The authors present evidence that image processing methods will prove most useful when analyzing these newer arrays. They state that with increasing use of microarrays by systems biologists, who would need to use a larger portion of the data, the importance of correcting for errors will increase. (Briefings in Bioinformatics, January 2008, pp. 25-33.)