EraGene: A Platform for Harnessing Recurrent Engineering Targets to Guide Metabolite Overproduction in Escherichia coli

EraGene is a platform that compiles experimentally validated engineering targets and production data from engineered Escherichia coli strains for metabolite overproduction. By integrating published metabolic engineering studies, it includes recurrently modified reactions, enabling identification of successful engineering strategies for biochemically related metabolites. We demonstrate its utility through two applications: identifying conserved engineering targets for pathway reinforcement and reducing the search space for constraint-based metabolic model predictions. Leave-one-out analysis showed that, on average, 76 ± 29% of a metabolite’s engineered reactions were already annotated for other metabolites. This overlap is statistically indistinguishable from a degree-preserving null model (p = 0.135) and is therefore reported as a descriptive measure of the operational redundancy of the curated corpus rather than as evidence of a privileged set of transferable targets. Incorporating reactions with conserved modulation direction into QPAML simulations reduced the average search space from 180.21 to 43.51 candidate reactions per metabolite while recovering 81.88% of experimentally validated modifications in the prediction set, corresponding to a 3.41-fold enrichment that remains significant against a null model preserving reaction popularity (p < 0.001). Independently, the communities detected in the knowledge graph were enriched for KEGG pathways well beyond a size-matched permutation null (20 versus 2.0 ± 1.3 enriched groups; p < 0.001). At publication, EraGene contained 3,833 production measurements from 2,611 strains across 360 studies, comprising 342 native reactions involving 921 genes. The core analytical dataset was restricted specifically to 300 native reactions associated with 88 target metabolites produced in E. coli using D-glucose as the sole carbon source. All analyses are retrospective, and the corpus records only successful, published interventions, so the reported predictive performance should be read as an upper bound pending prospective validation.