Systems Biology: genetic regulatory networks, structure, dynamics and inference

5 May 2006

Stuart Kauffman
The Institute for Biocomplexity and Informatics
The University of Calgary


Human cells have about 30,000 genes, each specifying the structure of one or a few related proteins. Different cell types, liver, kidney, spleen, differ because different genes are active in different cell types. These differences are largely due to the fact that the protein made by one gene can bind adjacent to another gene and turn it "on" or "off". Thus, there is a large genetic regulatory network with perhaps 2500 transcriptional factors, or proteins that regulate the activities of genes, regulating the 30,000 genes. This network is a complex non-linear dynamical system. The attractors of such networks are plausible models of cell types. However, at the single cell level the concentrations of individual molecular species are so low that the system is stochastic. Here "noisy attractors" are candidate models of cell types. In a class of deterministic model genetic networks, "Random Boolean nets", it has been shown that such networks behave in two regimes, ordered and chaotic, with a critical phase transition between them. Recent very tentative evidence suggests that real cells may be ordered or critical. However, the data are sparse, and it is not yet clear that these concepts apply to stochastic networks.

A central problem of Systems Biology is to einfer and confirm the structure and logic of genetic regulatory networks from data such as the time series of gene activities as cells differentiate from one cell type to another. We have developed an inference algorithm, IADGRN, for a specific class of model genetic regulatory networks, Random Boolean networks, (RBN), which surpasses known algorithms in its capacity to infer the structure and logic of noisy RBN. I will discuss this algorithm.

Real cells are stochastic, not Boolean networks. I will discuss an algorithmic approach to study the chemical master equation of model genetic networks, the Gillespie algorithm. We have already studied simple bistable networks. More importantly, I will discuss our plans to extend "Gillespie Networks" to the analysis of different classes, or ensembles, of model genetic networks, to seek general organizing principles underlying the dynamics of such networks, and our plans to extend IADGRN to Gillespie networks as a step towards the capacity to infer real genetic networks from time series of gene expression patterns.

If time permits, I will briefly describe our experimental program attempting to induce cancer stem cells to differentiate into normal cells.

current theory lunch schedule