Content area
Full Text
REVIEWS
Advantages and limitations of current network inference methods
Riet De Smet and Kathleen Marchal
Abstract | Network inference, which is the reconstruction of biological networks from high-throughput data, can provide valuable information about the regulation of gene expression in cells. However, it is an underdetermined problem, as the number of interactions that can be inferred exceeds the number of independent measurements. Different state-of-the-art tools for network inference use specific assumptions and simplifications to deal with underdetermination, and these influence the inferences. The outcome of network inference therefore varies between tools and can be highly complementary. Here we categorize the available tools according to the strategies that they use to deal with the problem of underdetermination. Such categorization allows an insight into why a certain tool is more appropriate for the specific research question or data set at hand.
The insight that genes and proteins do not work in isolation but act together in intricate networks has launched the era of systems biology1,2. In bacteria, regulation at the transcriptional level is pivotal to guaranteeing metabolic flexibility and cellular integrity1,2. In http://www.ncbi.nlm.nih.gov/sites/entrez?Db=genomeprj&cmd=ShowDetailView&TermToSearch=12319
Web End =Escherichia coli the transcription-regulatory network (TRN) was shown to be composed of basic modular components that contribute to the specificities of global response dynamics, for example by speeding up cellular responses or making them more robust (that is, able to respond to a wide range of environmental signals)3,4. Deciphering the gene co-expression network and the TRN (BOX 1)
is therefore crucial to understanding bacterial cellular behaviour. The number of computational methods that are being developed to reconstruct TRNs from genome-wide expression data is rapidly increasing; here, these methods are referred to as expression-centred methods. Module inference methods, which focus on the co-expression network, rely on the guilt-by-association
principle to identify functional relationships between genes, searching for gene sets or modules that exhibit a similar expression behaviour across experimental conditions (BOX 1). Methods that infer TRNs go one step beyond and infer causality relationships in the network by also identifying the transcriptional programmes of the genes or modules, to describe how transcription factors (TFs) cause the observed changes in expression of their cognate target genes (BOX 1).
Applying these inference procedures on public data sets of well-studied model organisms has considerably improved our...