Unsupervised Thresholding of Affymetrix Microarray Data

Unsupervised thresholding provides a data-driven alternative to manually-situated thresholds for those wishing to extract class structure from unlabelled data. The analysis of microarray data provides one such scenario, in which thresholds placed on the output of multiple hypothesis tests are the most common method of determining, for example, which genes of a genome-wide assay are expressed under different experimental conditions. The Affymetrix GeneChip microarray platform is a popular method of determining genome-wide gene expression. Here, we apply a well-known image segmentation algorithm to determine the simplest property inferred from Affymetrix microarray data $the detection of specific signal. The effective separation of specific and non-specific signal by an unsupervised thresholding algorithm demonstrates the potential of data-driven methods to complement and, in certain circumstances, replace manual thresholds in the analysis of this platform