An integrated knowledge-based approach for modelling biochemical reaction networks

This paper presents a new methodology for constructing cellular network topologies by searching for new binding species and new reactions catalysed by the enzymes present. Our technique is knowledge-based and integrates several steps. Starting from a pre-determined list of enzymes in the system it (i) generates lists of binding species, (ii) constructs a reaction network using these species and (iii) finds pathways through this network, which link different substrates (raw materials) with target metabolites (pathway products). Graph-theory-based analysis of the two-dimensional structures of known binding species is used to compute pharmacophores, the structures and functional groups binding at the corresponding enzymes’ sites. New binding species are obtained by searching in appropriate databases for existing compounds, which contain these pharmacophores. Reactions are constructed by generating all possible combinations of the binding species identified and by testing the feasibility (i.e. the ability to conserve atomic/molecular mass) of each constructed reaction. Generated reactions are required to be linearly independent in order to minimise the complexity of subsequent steps. Finally, pathways through the reaction network are computed to assess important reactions and metabolites for a given process. Our integrated procedure has been applied to two illustrative systems, the glycolysis and the citric acid cycle in Homo sapiens and Saccharomyces cerevisiae, respectively. New binding species and reactions were found for the enzymes involved. It was observed that some enzymes are very specific and only catalyse a small number of very similar reactions. Pathways were also constructed and analysed to demonstrate the relative importance of the metabolites involved.

► We build a new integrated procedure we call metabolic network development for constructing biochemical reaction networks and for finding pathways through them.
► We generate new binding species for sets of identified enzymes by computing the 2D pharmacophores (the maximum common substructures) of known binding species.
► We develop an automated procedure for generating new reactions based on sets of known binding species as well as new computed binding species.
► We construct new pathways for the sets of reactions generated which connect raw materials with pathway products.