Researchers from the University of Alberta, led by PhD student Fei Wang under the supervision of Russ Greiner and David Wishart, have successfully mapped the molecular dark matter of mammals. Published in Nature in January 2026, the study introduces a new artificial intelligence model named DeepMet, which predicts billions of previously unknown metabolites. The term dark matter in this chemical context refers to the vast number of small molecules found in bone and tissue that traditional mass spectrometry cannot identify. These unidentifiable compounds constitute the majority of the human metabolome, the complete set of chemical reactions and molecules within the body. DeepMet functions similarly to large language models that predict the next word in a sentence. However, instead of learning linguistic patterns, it analyzes the chemical structures of over 2,000 known human metabolites to understand the underlying logic of metabolism. By generating a billion potential molecular structures, the AI identified patterns that indicated biological viability. The researchers hypothesized that the structures appearing most frequently in the AI's output were the most likely to be real, existing metabolites. This approach allowed the team to anticipate biological chemistry and pinpoint the body's remaining unknown molecules. Since the release of the study, the team has validated their predictions by successfully identifying several dozen previously unrecognized mammalian metabolites in both human and mouse samples. This discovery holds significant potential for medical science. Metabolites are critical for bodily function, providing energy, building structures, and facilitating cellular signaling. Understanding these molecules offers a clearer picture of the body's current physiological state compared to genes, which act merely as a blueprint. According to the authors, a comprehensive map of the metabolome could lead to improved diagnostic tools and the development of new pharmaceutical drugs designed to target specific metabolic pathways more effectively. The project demonstrates how machine learning can solve complex biological problems that have long eluded traditional analytical methods. By treating metabolic chemistry as a language to be decoded, the researchers have opened a new frontier in understanding human biology. This breakthrough not only fills gaps in scientific knowledge regarding the mammalian metabolome but also sets a precedent for using AI to explore other areas of unknown chemistry. The study highlights the growing synergy between computing science, biochemistry, and artificial intelligence in addressing fundamental questions about life processes.