Classifying tumors

Microarray analysis has been used to find new subclasses in disease states and to identify biologic markers associated with disease. In a recent review, John Quackenbush of Dana-Farber Cancer Institute in Boston describes how microarray technology works and discusses whether patterns in gene expression profiles can be used to classify disease.

He wrote about a two-color and a single-color array for generating data and about confocal laser scanning for measuring the fluorescence intensity of each probe. Data are normalized to account for differences in labeling, hybridization and detection methods, filtered through a selected set of criteria, then represented in matrix form to indicate the levels of gene expression.

Among the ways to measure data in expression profiles, the author noted Pearson’s correlation coefficient distance for evaluating similarities in patterns, which are important in tumor classification. Data can be analyzed either by supervised or unsupervised methods, the latter of which disregard prior knowledge about the samples and are useful in identifying new subgroups. Hierarchical and k-means clustering are mentioned as commonly used methods.

Various methods of analysis can result in a variety of sets of significant genes. Although the best way to compensate for this is under active debate, Quackenbush presents several potential solutions, including publicizing data that meet industry standards. He also believes that gene selection criteria must be established and a method for validation developed.

Despite current limitations of classification, Quackenbush applauded the success of expression profiling with microarrays and said he believes that it will lead to advances in the diagnosis and treatment of disease. (The New England Journal of Medicine, June 8, 2006, p. 2463.)