Detection of phenotypes in microarray data using force- directed placement transformss

Abstract

Distinct cancer phenotypes are observable within a single type of cancer, corresponding to patients with different disease subtypes, prognosis, and treatment response. Analysis of correlations between genes and patients allows detection of gene sets that differentiate between these cancer phenotypes. We investigate the effect of force-directed placement transforms on bicluster-based feature selection for phenotype and marker detection. The transforms incorporate class-based metadata directly into the dataset topology and sharpen differences between classes. By incorporating important external clinical information such as disease status and using the transform for detection, the approach captures complex structure not visible from direct analysis of the data. When applied to model microarray data, the transform is shown to increase the quality of feature detection. On real microarray data, the transform offers higher sample enrichments and provides an alternative view of phenotypes not visible without the transform. © 2011 Springer-Verlag.