Microarrays—Identifying Molecular Portraits for Prostate Tumors with Different Gleason Patterns

We present in this chapter the combined use of several recently introduced methodologies for the analysis of microarray datasets. These computational techniques are varied in type and very powerful when combined. We have selected a prostate cancer dataset which is available in the public domain to allow for further comparisons with existing methods. The task is to identify biomarkers that correlate with the clinical phenotype of interest, i.e., Gleason patterns 3, 4, and 5. A supervised method, based on the mathematical formalism of (α,β)–k–feature sets (1 ) , is used to select differentially expressed genes. After these “molecular signatures” are identified, we applied an unsupervised method (a memetic algorithm ) to order the samples (2 ). The objective is to maximize a global measure of correlation in the two-dimensional display of gene expression profiles. With the resulting ordering and taxonomy we are able to identify samples that have been assigned a certain Gleason pattern, and have gene expression patterns different from most of the other samples in the group. We reiterate the approach to obtain molecular signatures that produce coherent patterns of gene expression in each of the three Gleason pattern groups, and we analyze the statistically significant patterns of gene expression that seem to be implicated in these different stages of disease.