Deep learning-driven automatic reconstruction of genome-scale metabolic networks

Abstract Incomplete knowledge of metabolic processes impairs the accuracy of GEnome-scale Metabolic models (GEMs), hindering advancements in systems biology and metabolic engineering. To close this critical gap, we present CLOSEgaps, a machine learning-based algorithm that considers the hypergraph topology of metabolic networks and hypothetical reactions to predict missing reactions and identify gaps in GEMs. Extensive results show that CLOSEgaps accurately gap-filled metabolic networks, filling over 96% of artificially introduced gaps, and enhances the predictability of fermentation products in 24 wild-type GEMs. Furthermore, we integrate CLOSEgaps into a generalized workflow for automated metabolic network reconstruction, hereby named NICEgame, and found a notable improvement in producing four crucial metabolites (Lactate, Ethanol, Propionate, and Succinate) in two organisms. As a broadly applicable solution for any GEM or reaction, CLOSEgaps promises to enhance biotechnological and biomedical applications by improving GEM-based predictions and automating the NICEgame workflow.