In Silico Evolution of Oscillating Mass-Action Chemical Reaction Networks

Abstract

Evolutionary algorithms, a class of optimization techniques inspired by biological evolution,have emerged as powerful tools for the optimization of complex systems, including the evolution of mass-action chemical reaction networks. This dissertation explores the application of evolutionary algorithms in this domain, presenting a novel approach inspired by neural network evolution methodologies. A key feature of the algorithm is speciation, separating candidate reaction networks into groups based on their similarity, which maintains diversity and protects innovations. Crossover has also been shown to be an effective means of improving evolutionary success in other domains. However, crossover of mass-action networks is tested and found to be detrimental to the evolutionary process. Other factors influencing evolutionary success are also explored and optimized resulting in a significant improvement over previous algorithms. This work goes beyond theoretical exploration by offering a practical contribution in the form of a user-friendly software module. This module encapsulates the newly devised algorithm, empowering researchers and practitioners to readily apply the speciation-based approach in their own investigations of mass-action chemical reaction networks.