A new AI system called MACROS can read a molecule's spectral fingerprint and identify its structure, working alongside chemists six times faster than either can alone.
Researchers built MACROS, a multi-agent AI system that automates organic structure elucidation, the process of figuring out a molecule's shape from spectroscopic data such as NMR. It was trained on 100 million simulated spectra-molecule pairs plus 1.6 million real experimental ones, and it mimics how expert chemists work, forming a hypothesis, testing it against the data, and revising until it fits. In tests on real-world samples it had never seen, MACROS correctly identified synthetic compounds, natural products, and metabolites heavier than 500 daltons using only 1D NMR data. The system also recovered textbook spectroscopic correlations on its own and developed habits such as reading ring structures first, a sign it learned actual chemistry rather than memorizing a database.
Structure elucidation has long been a slow, manual bottleneck in drug discovery and chemical research, even for experienced chemists. Paired with a human chemist, MACROS delivers results six times faster and 40% more accurate than either the AI or the chemist working alone. That gain comes from collaboration, not full automation.
It is a tool meant to scale chemists' judgment, not replace it, though the researchers frame it as a step toward fully autonomous laboratories.