Meta-analysis on gene regulatory networks discovered by pairwise Granger causality

Abstract

Identifying regulatory genes partaking in disease development is important to medical advances. Since gene expression data of multiple experiments exist, combining results from multiple gene regulatory network discoveries offers higher sensitivity and specificity. However, data for multiple experiments on the same problem may not possess the same set of genes, and hence many existing combining methods are not applicable. In this paper, we approach this problem using a number of meta-analysis methods and compare their performances. Simulation results show that vote counting is outperformed by methods belonging to the Fisher's chi-square (FCS) family, of which FCS test is the best. Applying FCS test to the real human HeLa cell-cycle dataset, degree distributions of the combined network is obtained and compared with previous works. Consulting the BioGRID database reveals the biological relevance of gene regulatory networks discovered using the proposed method.

Novel algorithms for inferring granger causality using biological data

“…Moreover, the statistical concepts behind the estimation of VAR models using observational data and its computational algorithms are well understood. Some VAR models have been employed successfully in several areas, such as economics [21,42,43], neuroscience [44][45][46][47][48], and, more recently, system biology [29,[49][50][51][52][53].…”

“…Pairwise Granger causality is an extension of standard OLS-based Granger causality analysis, as discussed in Chapter 2, and was suggested by Tam et al [51,114]. They proposed that for high-dimensional data, we can analyse variables in pairs and combine all the resultant matrices to form a single matrix, thereby by-passing the OLS limitation of handling high-dimensional data (i.e., when the number of variables is greater than the number of observational time points).…”