Correlation and Gene Co-Expression Networks

Abstract

A correlation network is a network whose adjacency matrix is constructed on the basis of pairwise correlations between numeric vectors. The numeric vectors may represent observed quantitative measurements of variables. For example, the gene expression levels (transcript abundances) across different conditions can be represented by a numeric vector. In general, the relationship between a pair of numeric vectors can be measured in many ways, in particular, using a correlation coefficient (e.g., the Pearson-, Spearman-, or biweight mid-correlation) or using the concordance index. Mouse gene expression data are used to illustrate how network concepts can be used to describe the pairwise relationships among gene expression profiles. While cluster trees and heat maps can be used to visualize relationships between variables, concepts of correlation networks can be used to quantify them. Brain cancer gene expression data are used to illustrate the topological effects of hard- and soft-thresholding. We provide an overview of weighted gene coexpression network analysis and different gene network (re-)construction methods.