Toggle Main Menu Toggle Search


Inference of chemical reaction networks using hybrid s-system models

Lookup NU author(s): Dr Dominic Searson, Dr Mark Willis, Simon Horne, Professor Allen Wright


Full text for this publication is not currently held within this repository. Alternative links are provided below where available.


This article demonstrates, using simulations, the potential of the S-system formalism for the inference of unknown chemical reaction networks from simple experimental data, such as that typically obtained from laboratory scale reaction vessels. Virtually no prior knowledge of the products and reactants is assumed. S-systems are a power law formalism for the canonical approximate representation of dynamic non-linear systems. This formalism has the useful property that the structure of a network is dictated only by the values of the power law parameters. This means that network inference problems (e.g. inference of the topology of a chemical reaction network) can be recast as parameter estimation problems. The use of S-systems for network inference from data has been reported in a number of biological fields, including metabolic pathway analysis and the inference of gene regulatory networks. Here, the methodology is adapted for use as a hybrid modelling tool to facilitate the reverse engineering of chemical reaction networks using time series concentration data from fed-batch reactor experiments. The principle of the approach is demonstrated with noisy simulated data from fed-batch reactor experiments using a hypothetical reaction network comprising 5 chemical species involved in 4 parallel reactions. A co-evolutionary algorithm is employed to evolve the structure and the parameter values of the S-system equations concurrently. The S-system equations are then interpreted in order to construct a network diagram that accurately reflects the underlying chemical reaction network.

Publication metadata

Author(s): Searson DP, Willis MJ, Horne SJ, Wright AR

Publication type: Article

Publication status: Published

Journal: Chemical Product and Process Modeling

Year: 2007

Volume: 2

Issue: 1

Print publication date: 01/05/2007

ISSN (print): 1934-2659

Publisher: Berkeley Electronic Press


Notes: Software tools and algorithms that can reliably reverse engineer one or more plausible chemical reaction networks from experimental laboratory data are the subject of considerable interest. For instance, a key piece of information within the development lifecycle of new chemical entities (NCEs) is a kinetic model of the reaction system. Once obtained, this allows the chemical engineer to understand the underlying chemical processes and facilitates the use of reaction engineering, modelling and simulation tools. Commercial software exists for kinetic fitting, once the structure of the reaction network is known, but the initial elucidation of this structure – a network of reactants and products – is a significant bottleneck within the development cycle that requires the expertise of both chemists and engineers. Methods that can accelerate the discovery of chemical reaction networks, using minimal prior information, have substantial commercial and academic potential. In particular, the advent and continuing development of high throughput technologies (HTT), e.g. automated robotic workstations for performing many experiments in parallel, coupled with improved chemical sensor technology will provide an increase in the quantity and quality of experimental data available during the NCE lifecycle. This trend suggests that effective methods for the reverse engineering of reaction networks from data will become of increasing importance in the future. In this paper, an S-system methodology for elucidating the topology of an unknown reaction network - using time series measurements made from fed-batch reactors - is described. When combined with a priori information, e.g. knowledge of the relative molecular masses of the reacting species, the topology of the S-system model can allow a set of consistent mechanistic reaction steps to be deduced.


    Link to this publication