To extract useful biological information from DNA microarray experiments, it is necessary to accurately quantify the measured expression levels and to systematically organize them in biologically meaningful ways. There are many options for the implementation of feature extraction and gene clustering programs to help accomplish these goals. We have developed Matlab prototypes that emphasize flexibility, accuracy, transparency and the systematic incorporation of statistical analysis. These prototypes facilitate the development and refinement of algorithms as we converge on fully automatic analysis.
There are many possible approaches to feature extraction, and the choices are influenced by factors including: feature morphology and uniformity, and positioning errors, array deposition methodologies and tolerances, labelling methods and scanner specifications. Our feature extraction software includes several methodologies and a number of adjustable parameters that can be modified to suit the application. For example, image file formats, image alignment procedures for two-colour images, irregular modifiable grids, feature locating, grid fitting, extraction, pixel-level outlier rejection, feature-level outlier rejection, background-subtraction and dye-normalization processes have all been incorporated. With this feature extraction prototype, we are can explore different methods of image analysis, outlier rejection, background subtraction and normalization while optimizing sensitivity and specificity. Relative expression levels are reported along with their respective statistical information.
This is a preview of subscription content, access via your institution