DeepMet is a chemical language model focused on the human metabolome, the collection of metabolites associated with humans. It represents chemical structures as concise text strings, enabling neural network methods commonly used to model language to analyze atoms and bonds. By learning patterns in the structures of known human metabolites, DeepMet aims to capture aspects of their underlying biosynthetic logic and use that knowledge to predict metabolites that may not yet have been discovered.

The DeepMet website provides tools for exploring compounds and conducting chemical searches. Visitors can browse compounds and use mass search, formula search, and MS/MS search to investigate chemical data. The site also offers information about the DeepMet approach and its use of language-model techniques for chemical structures. Together, these resources support research and exploration of known and predicted human metabolites, connecting computational modeling with practical ways to examine compounds. DeepMet’s focus is on applying learned structural patterns to metabolite discovery, rather than presenting a general-purpose chemical catalog or consumer product.