Reverse Engineering of Biochemical Reaction Networks Using Co-evolution with Eng-Genes

A major challenge when attempting to model biochemical reaction networks within the cell is that the dimensionality can become huge, where a large number of molecular species can be involved even in relatively small networks. This investigation attempts to infer models of these networks using a co-evolutionary algorithm that reverse engineers differential equation models of the target system from time-series data. The algorithm not only estimates the system parameters, but also the symbolic structure of the network. To reduce the problem of dimensionality, the algorithm uses a partitioning method while integrating candidate models in order to decouple system equations. In addition, the conventional evolutionary algorithm has been modified and extended to include a technique called 'eng-genes', where candidate models are built up from fundamental mathematical terms derived from knowledge about the target system a priori. This technique essentially focuses the search on more biologically plausible models. The approach is demonstrated on several example reaction networks. The results show that the eng-genes method of limiting the term pool using a priori knowledge improves the convergence of the reverse engineering process compared with the conventional method, resulting in more accurate and transparent models.