The availability of different levels of omics data helps us to observe cells with higher resolution and from different perspectives. However, complete understanding of metabolism lags behind in explaining the chemodiversity observed in living organisms – the known reactome does not account for the appearance of many metabolites. We applied the computational tool BNICE.ch to build the “ATLAS of Biochemistry” using expert curated generalized enzyme reaction rules. The first version of ATLAS contains all possible reactions (known and hypothetical) between known biological compounds. In the extended version of ATLAS, we extrapolate the known metabolism towards the chemical knowledge space and we integrate in a supervised manner chemical compounds and their associated reactions into a super network of metabolism. We demonstrate that the supervised integration of chemicals into metabolic networks is the key to complete the mechanism of poorly characterized reactions and to integrate thousands of orphan metabolites into metabolic networks. ATLAS is available as an online database (http://lcsb-databases.epfl.ch/atlas) and is equipped with additional data analysis tools. ATLAS can further be used to create hypotheses about the origin of experimentally measured compounds and serve as a tool for metabolic engineers, synthetic biologists and other scientists working with metabolomics and secondary metabolism.