Invisible "dark matter" - what cosmologists call the mysterious glue that holds everything together - is estimated to make up more than a quarter of the universe.
In chemistry, dark matter refers to the thousands of small molecules in bone and tissue that can't be identified using mass spectrometry. They make up the vast majority of what's known as metabolites.
Now, University of Alberta computing science PhD student Fei Wang, under the supervision of computing scientist Russ Greiner and biochemist David Wishart, and an international research team have come up with a "chemical language model" trained on the chemical structures of more than 2,000 known human metabolites to map the dark matter of the entire mammalian metabolome. The results of their discovery were published in Nature in January.
In much the same way artificial intelligence models like ChatGPT learn language patterns to predict the next word in a sentence, DeepMet learns the "logic" of metabolism from the chemical structures of known metabolites to predict those not yet known.
By generating a billion potential structures, the researchers speculated that molecules the AI created most often were the most likely to be real metabolites, essentially anticipating biological chemistry to identify the body's remaining unknown molecules. The team has so far successfully identified several dozen previously unrecognized mammalian metabolites in human and mouse samples.
Metabolites are a key part of what allows a body to function, so knowing how they provide energy, build structures and send signals to different parts of the body could lead to better diagnostics or new pharmaceutical drugs that target metabolic pathways more effectively, say the study's authors.
"Genes are like the blueprint for your body, describing what you were born with. Metabolites are what's happening in your body right now, which is clearly important," Greiner says.
