Until recently, the large-scale modeling of network dynamics has been focused on individual network types. However, within a cell, all network types are interrelated and dynamics of any individual network has an impact on the behavior of other networks. Several recent studies have begun to address the challenge of coupling large-scale dynamical models for different network types to obtain one consistent dynamical network. Such methods have been spearheaded by approaches to combine metabolic and regulatory networks (see [6–8] and references therein). For example, to obtain a combined model of metabolic and regulatory networks, Covert et al. [6] used flux-balance analysis to model the metabolic network component while the transcriptional regulatory network was modeled as a Boolean network. The genes in the transcriptional network were assigned Boolean (binary) values indicating whether or not a given gene is being expressed. An interactive procedure was applied to ensure that the combined model satisfies both the metabolic and the regulatory constraints. A subsequent study used mixed integer linear programming (a general optimization framework for capturing problems with both discrete and continuous variables) to couple such metabolic and regulatory models [8].
In their recent paper in BMC Biology, Wang and Chen [9] propose a promising approach for integrating transcription regulation and protein-protein interactions using dynamic gene-expression data. They start with candidate gene regulatory and signaling networks obtained from genome-scale data. These candidate networks are then pruned and combined, utilizing gene-expression data at multiple time points, to obtain an integrated and focused network under a specific condition of interest. The transcriptional network is modeled as a dynamical system in which the expression of a target gene (a gene subject to regulation by transcription factors included in the network) is computed as a function of regulatory impact of the corresponding transaction factors, its expression at a previous time point, and mRNA degradation rate. The modeling of a signaling/protein-interaction network takes into account, among other factors, the activities of its neighbors in the network. The interaction rate between two neighboring proteins is assumed to be proportional to the product of their concentrations. An overview of the method used by Wang and Chen [9] is depicted in Figure 1 and further details are given in Figure 2.
Wang and Chen applied their method to Saccharomyces cerevisiae (budding yeast) networks for three different stress responses - hyperosmotic stress, heat-shock stress and oxidative stress - and identified highly connected transcription factors and genes. Further analysis of the crosstalk between these three networks revealed the significance of some transcription factors in serving as the decision-making devices and in playing a role in rapid adaptation in the stress-response mechanism.
The authors also showed that their method can be used to predict gene-expression levels under different conditions. To do so, they first constructed the integrated network under heat-shock stress for the wild-type strain of yeast and then used the trained data to predict the expression level of the gene HXT5 in the yap1 mutant strain, which had been originally determined by Gasch et al. [10]. Their results suggest that various types of network models can be combined successfully to yield a predictive dynamic model of the heterogeneous system.