Evolving Chemical Reaction Networks

Sammanfattning: One goal of synthetic biology is to implement useful functions with biochemical reactions, either by reprogramming living cells or programming artificial vesicles. In this perspective, we consider Chemical Reaction Networks (CRNs) as a programming language. Recent work has shown that continuous CRNs with their dynamics described by ordinary differential equations are Turing complete. That means that any function over the reals that is computable by a Turing machine in arbitrary precision, can be computed by a CRN over a finite set of molecular species. The proof uses an algorithm which, given a computable function presented as the solution of a PIVP (PolynomialInitial Values Problem), generates a finite CRN to implement it. In the generated CRNs, the molecular concentrations play the role of information carriers, similarly to proteins in cells. In this Master’s Thesis, we investigate an approach based on an evolutionary algorithm to build a continuous CRN that approximates a real function given a finite set of the values of the function. The idea is to use a two-level parallel genetic algorithm. A first algorithm is used to evolve the structure of the network, while the other one enables us to optimize the parameters of the CRNs at each step. We compare the CRNs generated by our method on different functions. The CRNs found by evolution often give good results with quite unexpected solutions.