Reconstructing Chemical Reaction Networks: Data Mining Meets System Identification

Overview

This paper presents an approach to reconstructing chemical reaction networks from time series measurements of the concentrations of the molecules involved. The solution strategy combines techniques from numerical sensitivity analysis and probabilistic graphical models. By modeling a chemical re-action system as a Markov network (undirected graphical model), the paper shows how systematically probing for sensitivities between molecular species can identify the topology of the network. Given the topology, the approach next uses detailed sensitivity profiles to characterize properties of reactions such as reversibility, enzyme-catalysis, and the precise stoichiometries of the reactants and products. The paper demonstrates applications to reconstructing key biological systems including the yeast cell cycle. In addition to network reconstruction, the algorithm finds applications in model reduction and model comprehension.