Exploring Gene Switches: a Mining Approach to Predict Dynamic Features of Gene Network Based on Expression Data

An important topic in systems biology is to understand the dynamic properties of biological networks. One of the dynamic features that emerges from genetic regulatory network is the “switch-like behavior” or “bistability”, a non-linear response characterized by two distinct steady states. Gene switches, or genetic switches have been reported to play pivotal roles in many important aspects of cell physiology, cell fate decisions, cell cycle control and cellular responses to environmental stimulation.

Previous engineering studies have provided important insights into the mechanistic principles of switch-like behaviors, while in biological system, only a couple of studies thus far have demonstrated functional roles of gene switches by kinetic modeling and simulation. Thus far, current methods have been limited to well-characterized realistic or synthetic small modules. However, in most cases a detailed kinetic model is not available for biological processes due to the difficulties in quantitatively measuring the network components and reaction constants. Since switches usually play a central role in cell decision, it is beneficial to be able to predict whether switches can occur without knowing, a priori, all the details of the network. The essential problem is how to bridge such dynamic behaviors described by bottom-up mechanisms with top-down global functional analysis, i.e. to associate the dynamic features of single molecules with the large-scale network structure and functionality.

In this study, we propose a top-down mining approach to explore genetic switches on a genome-scale level. Taking advantage of the tremendous amount of expression data, our approach aims to identify bimodality, which we believe is an essential characteristic of a gene switch. We perform theoretical analysis and provide proof-of-concept applications on both synthetic and yeast microarray datasets. We test our hypothesis that the dynamic property of molecular components within a network is contingent on the network organization and specific biological functionality. We further apply our system on an integrated human expression dataset comparing multiple cancer cell types and non-cancer cases to probe characteristic signatures of human cancer. These characteristics may govern the switch-like behavior and thereby determine the status of the biological systems and whether it could be perturbed to a cancerous phenotype. To the best of our knowledge, this is the first attempt in applying mining approaches to explore genetic switches on a genomic-scale.