StereoMapper: Clarifying Metabolite Identity Through Stereochemically Aware Relationship Assignment
Inaccuracies in metabolite cross-mapping frequently introduce errors into genome-scale metabolic models (GEMs). A key challenge lies in distinguishing truly identical metabolites from closely related structural variants, such as stereoisomers, across biochemical databases. StereoMapper addresses this issue by establishing stereochemically aware relationships that clarify metabolite identity. It operates entirely on structural information, using molecular structures to infer equivalence and define relationships. StereoMapper was implemented using RDKit (2025.3.3) and Open Babel-wheel (3.1.1.22) which provides the core functionality of the Open Babel software, and benchmarked across four curated control datasets to evaluate performance in relationship classification (enantiomer, diastereomer, stereo-resolution pairs, and protomer). The pipeline was subsequently applied to 1.3 million molecular structures from multiple metabolic databases to generate equivalent isomeric sets (groups of structures identified as equivalent by StereoMapper). These sets were compared against the human subset of MetaNetX, a context-aware reference, to assess concordance and investigate disagreements. On curated positive control sets containing only positive examples across multiple relationship classes, StereoMapper achieved mean precision, recall, and F1-scores of 92.6, 98.0, and 95.3, respectively; following optimisation, these improved to 98.9, 98.2, and 98.5. Application to the 1.3 million structures produced over 339,000 stereochemically defined relationships between molecular structures. Comparison with the human subset of MetaNetX assignments showed an overall grouping concordance of 93.5%. The remaining 6.5% of differences were mainly associated with protomer relationships (52.3%) and stereo-resolution pairs (37.4%), while 21.6% involved complex or ambiguous cases. Scientific contribution statementStereoMapper provides a stereochemically aware pipeline for metabolite relationship assignment that shows high performance across multiple benchmark datasets, resolving key ambiguities in metabolite identity and providing a complementary approach to existing context-aware frameworks. Its integration will enhance reaction-level cross-mapping and improve the accuracy of future GEM reconstructions.