Median correlation for the analysis of gene expression data

Microarray technology is revolutionizing functional genomics research by allowing scientists to measure the expression level of thousands of genes simultaneously from a single sample. However, a standard protocol for microarray data analysis has yet to be established. Many analysis techniques currently rely on linear correlation between pairs of genes. Such analysis only detects relationships in signal components exhibiting Gaussian correlation statistics. This paper focuses on determining the relationships between gene patterns based on a hierarchical clustering of nonlinear correlation measurements. The methods described herein are illustrated with gene expression data from yeast, and from human cancer cell lines. The results indicate that in some cases, improved clustering of genes can be achieved by the use of nonlinear correlation metrics. (C) 2002 Elsevier Science B.V. All rights reserved.