Reactant pairs and reaction organization patterns produced by a new rule-based approach.

OBJECTIVES: Improvements in bioinformatics applications for the enzyme identification of biochemical reactions, enzyme classifications, mining for specific inhibitors and pathfinding require the accurate computational detection of reaction similarity. We provide a set of substrate-product pairs, clustered by reactions that share similar chemical transformation patterns, for which accuracy was calculated, comparing this set with manually curated data sets. DATA DESCRIPTION: The data were analyzed by a new method that naturally split each reaction into compound pairs and loner compounds, which we called architectures (Vazquez-Hernandez et al. in BMC Syst Biol 12:63, 2018). The data include a set of 7491 curated reactions from the KEGG-Ligand data set. The data are presented in two formats, a string format and a tree structure, both of which reflect the splitting process and the final architectures of each reaction. We are also reporting sets of reactions that show similar splitting patterns naturally grouped into clusters of tree structures. The compound pairs in each cluster were compared with the reactant pairs proposed by the KEGG-RCLASS data set, and a match precision value is also provided. These data were collected with the aim of providing research with a confident set of reactant pairs that is useful for selecting between alternative substrate-product pairs predicted by pathfinders.

Identification of reaction organization patterns that naturally cluster enzymatic transformations.

BACKGROUND: Metabolic reactions are chemical transformations commonly catalyzed by enzymes. In recent years, the explosion of genomic data and individual experimental characterizations have contributed to the construction of databases and methodologies for the analysis of metabolic networks. Some methodologies based on graph theory organize compound networks into metabolic functional categories without preserving biochemical pathways. Other methods based on chemical group exchange and atom flow trace the conversion of substrates into products in detail, which is useful for inferring metabolic pathways. METHODS: Here, we present a novel rule-based approach incorporating both methods that decomposes each reaction into architectures of compound pairs and loner compounds that can be organized into tree structures. We compared the tree structure-compound pairs to those reported in the KEGG-RPAIR dataset and obtained a match precision of 81%. The generated tree structures naturally clustered all reactions into general reaction patterns of compounds with similar chemical transformations. The match precision of each cluster was calculated and used to suggest reactant-pairs for which manual curation can be avoided because this is the main goal of the method. We evaluated catalytic processes in the clusters based on Enzyme Commission categories that revealed preferential use of enzyme classes. CONCLUSIONS: We demonstrate that the application of simple rules can enable the identification of reaction patterns reflecting metabolic reactions that transform substrates into products and the types of catalysis involved in these transformations. Our rule-based approach can be incorporated as the input in pathfinders or as a tool for the construction of reaction classifiers, indicating its usefulness for predicting enzyme catalysis.