Section: New Results

Structural identification of gene regulatory networks

In general, structural identification of genetic regulatory networks involves fitting appropriate network structures and parameters to the data. While modern measurement techniques such as reporter gene systems provide data of ever-increasing quality, the problem remains challenging because exploring all possible network structures in the search of the best fitting model is prohibitive.

In order to address the structural identification problem, Eugenio Cinquemani developed in collaboration with the Automatic Control Lab at ETH Zürich (Switzerland) and the Computer Engineering & Systems Science Department of the University of Pavia (Italy), an ODE modelling framework based on so-called unate-like functions, and a method that exploits monotonicity properties of these functions to effectively prune models that are incompatible with the data from the family of all unate-like modelling alternatives. This model invalidation step is based on simple preprocessing of time-course protein concentration and synthesis rate profiles, assumed available, and allows one to reduce the search of the best fitting model to a small subset of viable model structures.

The method, first published in Bioinformatics in 2010 and demonstrated on real data from the synthetic network IRMA, allows one to integrate a-priori knowledge on the expected network dynamics in a natural way. Leveraging on this, in the context of the same international collaboration, the method has been further developed in particular by considering relevant subclasses of the family of unate-like models that also enjoy certain quasi-convexity properties. For this restricted class, combined use of monotonicity and quasi-convexity properties allows one to ameliorate the model invalidation step, i.e. retain even fewer viable model structures based on affordable data preprocessing. These developments have been presented and demonstrated in silico in a paper published in the 2012 special issue on System Identification for Biological Systems of the International Journal of Robust and Nonlinear Control [9] .

We are currently applying the above methods to actual, known or partially unknown, networks. In the framework of the PhD thesis of Diana Stefan, the network inference method has been applied to gene expression data from the network regulating motility of E. coli. First encouraging results have suggested further experimental and computational investigations that are currently in progress.