A comparative study of statistical methods used to identify dependencies between gene expression signals

Abstract

One major task in molecular biology is to understand the dependency among genes to model gene regulatory networks. Pearson's correlation is the most common method used to measure dependence between gene expression signals, but it works well only when data are linearly associated. For other types of association, such as non-linear or non-functional relationships, methods based on the concepts of rank correlation and information theory-based measures are more adequate than the Pearson's correlation, but are less used in applications, most probably because of a lack of clear guidelines for their use. This work seeks to summarize the main methods (Pearson's, Spearman's and Kendall's correlations; distance correlation; Hoeffding's D: measure; Heller-Heller-Gorfine measure; mutual information and maximal information coefficient) used to identify dependency between random variables, especially gene expression data, and also to evaluate the strengths and limitations of each method. Systematic Monte Carlo simulation analyses ranging from sample size, local dependence and linear/non-linear and also non-functional relationships are shown. Moreover, comparisons in actual gene expression data are carried out. Finally, we provide a suggestive list of methods that can be used for each type of data set.

Selecting biologically informative genes in co-expression networks with a centrality score

“…A wide diversity of measures are available, and previous research has shown the pros and cons of each option in different applications [6,46,47]. Deeper investigations of the effects of measure choice on the structure and function of co-expression networks is warranted.…”

Differential correlation for sequencing data

“…Efforts to find a single best metric to measure correlation for –omics data is inconclusive because performance depends on the distribution of the data, sample size and the observation of interest [13]. For example, Pearson’s correlation can be used when the distribution is approximately Gaussian and the user has interest in linear relationships, while Hoeffing’s D measure may be better suited for non-monotonic associations [13].…”

“…For example, Pearson’s correlation can be used when the distribution is approximately Gaussian and the user has interest in linear relationships, while Hoeffing’s D measure may be better suited for non-monotonic associations [13]. Two studies compared correlation statistics for –omics data but were validated using microarrays [13, 14].…”

“…For example, Pearson’s correlation can be used when the distribution is approximately Gaussian and the user has interest in linear relationships, while Hoeffing’s D measure may be better suited for non-monotonic associations [13]. Two studies compared correlation statistics for –omics data but were validated using microarrays [13, 14]. In our work, we examine a variety of correlation statistics appropriate for sequencing data for the purpose of testing DC including Pearson’s, Spearman’s, biweight midcorrelation (BWMC) and a sparse compositional correlation method SparCC [15].…”

Gene Networks in Plant Biology: Approaches in Reconstruction and Analysis

“…Another widely used coefficient is mutual information (MI) [29] which is based on information theory. Some studies have shown that MI is able to detect nonlinear and nonmonotonic relationships [30], but other studies have suggested that MI does not outperform Box 2. Determining the Sample Size…”