Measurement variation determines the gene network topology reconstructed from experimental data: a case study of the yeast cyclin network.

Inference of the topology of gene regulatory networks from experimental data is one of the primary challenges of systems biology. In an example of a genetic network of cyclins in the yeast cell cycle, we analyzed static genome-wide location data together with microarray kinetic measurements using a recurrent neural network-based model of gene expression and a newly developed, unbiased algorithm based on evolutionary programming principles. The modeling and simulation of gene expression dynamics identified cyclin genetic networks that were active during the cell cycle. We document that because there is inherent experimental variation, it is not possible to identify a single genetic network, only a set of equivalent networks with the same probability of occurrence. Analysis of these networks showed that each target gene was controlled by only a few regulators and that the control was robust. These results led to the reformulation of the cyclin genetic network in the yeast cell cycle as previously published. The analysis shows that with the methodologies that are currently available, it is not possible to predict only one genetic network; rather, we must work with the hypothesis of multiple, equivalent networks. Chromatin immunoprecipitation (ChIP)-on-chip experiments are not sufficient to predict the functional networks that are active during an investigated process. Such predictions must be considered as only potential, and their actual realization during particular cellular processes must be identified by incorporating both kinetic and other types of data.