Automatic reconstruction of the mouse segmentation network from an experimental evidence database

Mammalian vertebrae, ribs, body wall musculature and back skin develop from repetitive embryonic tissues called somites. The development of somites depends on the molecular oscillations of the products of so-called cyclic genes. The underlying network involves the Wnt, Fgf/Mapk, Notch signaling pathways and the T-box genes. The discovery of this network is based on genetic interactions. Because of regulatory feedbacks and cross-regulation between pathways, it is often difficult to intuitively identify direct molecular interactions underlying genetic interactions. To address this problem, we developed a method based on a database and graph theory algorithms. We first encoded genetic and non-genetic experiments in a relational database. Next, we built a reference network with the data from non-genetic experiments and the KEGG pathway database. Then, we computed the shortest path between the nodes for each genetic interaction in the reference network to propose direct molecular interactions. The resulting network is the largest computational representation of the mammalian segmentation network to date with 36 nodes and 57 interactions. In some instances, a number of genetic interactions could be explained by adding a single link to the reference network, which leads to experimentally testable hypotheses. Two examples of such predictions are the direct transcriptional regulation of Dll3 and Fgf8 genes by the Rbpj and Ctnnb1 products, respectively. Furthermore, the computed shortest paths suggest that cross-talks from the Wnt to the Fgf/Mapk and Notch pathways might be mediated by the Dvl genes. This method can be applied in any system where gene expression changes are observed as a response to some gene perturbation, for instance in cancer cells.