Reconciliation of metabolites and biochemical reactions for metabolic networks

Abstract

Genome-scale metabolic network reconstructions are now routinely used in the study of metabolic pathways, their evolution and design. The development of such reconstructions involves the integration of information on reactions and metabolites from the scientific literature as well as public databases and existing genome-scale metabolic models. The reconciliation of discrepancies between data from these sources generally requires significant manual curation, which constitutes a major obstacle in efforts to develop and apply genome-scale metabolic network reconstructions. In this work, we discuss some of the major difficulties encountered in the mapping and reconciliation of metabolic resources and review three recent initiatives that aim to accelerate this process, namely BKM-react, MetRxn and MNXref (presented in this article). Each of these resources provides a pre-compiled reconciliation of many of the most commonly used metabolic resources. By reducing the time required for manual curation of metabolite and reaction discrepancies, these resources aim to accelerate the development and application of high-quality genome-scale metabolic network reconstructions and models.

Keywords: cheminformatics; data integration; data interoperability; metabolic networks; metabolic resources.

Figure 1:

Examples of the types of problems that are frequently encountered when attempting to reconcile metabolite representations from different resources.

Figure 2:

Different types of string representations of the structure of β-d-glucose 6-phosphate(2−). Different SMILES can be defined for the same chemical structure (only a few have been reported here). Both generic SMILES and isomeric SMILES describe atom connectivity, but only isomeric SMILES encode the stereo-specificity. The standard InChI is unique to a structure and describes distinct aspects of chemical compound structure in distinct ‘layers’. This architecture allows the comparison of compounds at different levels of ambiguity. The InChI key encodes the information contained in the InChI in a more compact way, facilitating integration and comparison of InChIs.

Figure 3:

Example of the reconciliation of 2-methylcitrate, as performed within MNXref. The reconciliation is performed in two main steps: Stage 1: reconciliation of the metabolites using structural information. At Stage 1.1, the InChI is computed for the major tautomeric form of each compound with a structure using ChemAxon software. The choice of pH 7.3 is arbitrary and in line with resources such as MetaCyc and Rhea. Compounds with identical InChI at pH 7.3 correspond to different protonation states or tautomeric forms of the same metabolite and are considered unique. InChI removes metal bonds by default, eliminating difficulties linked to conventions used in the representation of organometallic complexes. At Stage 1.2, MNXref uses the following heuristic to disambiguate true different isomeric metabolites from incomplete knowledge. We take advantage of the information present in all public reactions databases. If none of these reaction databases use two different stereoisomeric forms of the same molecule, then we assume that there is currently no reason to make a distinction between the different stereochemical forms of this metabolite, and merge them. Otherwise we keep them as independent entities. Stage 2: describes the reconciliation of metabolites lacking structural information using reaction context (which is detailed in Figure 4).

Figure 4:

Principle of the reconciliation of metabolites using reactions context (MNXref). Because structural information is lacking for some compounds, the MNXref reconciliation process attempts to infer links between compounds through the reaction context. Reactions are paired if they share all reactants but one or if they have been paired by reaction cross-references. Thereafter, the possible mappings between compounds are exhaustively enumerated, and conflicting mappings are resolved by a majority vote rule. Finally a mapping is accepted only if the chemical formulae and charges of the two compounds match (they must correspond to the same molecule regardless pH) and if they have a common name or synonym. This procedure is iterated until no new mappings can be obtained.

Figure 5:

Comparison of the reconciliations of BKM and MNXref. The Venn diagrams show the reconciliation of compounds and reactions from BRENDA, KEGG and MetaCyc, the three resources that are common to both BKM-react and MNXref. The number of compounds and reactions contributed by each of the three resources is indicated in parentheses for each. As MNXref contains data from more recent releases than BKM-react, we have included only the common subset of compounds and reactions in the comparison.