A Review of Integration Strategies to Support Gene Regulatory Network Construction

Figure 5

Workflow of the algorithm by Segal et al. This is an iterative procedure with application of the expectation maximization (EM) algorithm. In the maximization step (M step), genes are partitioned into modules that result from previous clustering upon genomic expression data and the best regulation program is learned for each module. In the E step, the best regulation programs corresponding are compared with each gene module to determine the optimal predictor (the optimal predictive regulation program). The module corresponding to the best predictor is selected and genes are reassigned to this module. The regulatory program learning stops on convergence. Secondly, TFs are associated with the regulatory module via an enrichment test of their corresponding binding motifs to the module.