Reconstruction of metabolic and genetic networks from gene expression perturbation data using a Boolean model: Construction of a simulation testbed and an empirical exploration of some of the limits

High-throughput molecular biology techniques—in particular, gene expression microarrays—are now producing data in quantities large enough for researchers to attempt reconstructions of metabolic and genetic networks primarily based on such data. This work explores the use of Boolean models in reconstruction of the topology of such networks. The construction and employment of a software suite for such exploration is described. The program suite forms a testbed for reconstructions of the regulatory edges of simulated networks of different types, using a Boolean model for the gene expression values and the node states in the networks. Using gene expression data from simulated perturbations, the relative difficulty of reconstruction of different networks is measured. Important network parameters are determined. Target in-degree is found to be the most important variable. Also, the effects of noise (random errors) in the gene expression measurements are described. Also, different inference methods are compared against the same networks, for measurement of their relative power. The value of control points into the networks (settable inputs into the nodes) is described. The testbed is used to refine one of the original inference methods, conditional mutual information inference (CMI), doubling its power in terms of the target in-degree it can handle. This refinement allows near-perfect reconstruction using CMI of the genetic networks tested with target in-degree of two or less that use input switches, when done in the absence of noise. Such reconstruction requires a very small number of random perturbations relative to the space of all possible perturbations.