MACROS: Multi-Agent AI System Revolutionizes Organic Structure Elucidation from Spectra
The newly created multi-agent AI system, MACROS, aims to automate the elucidation of organic structures from standard spectroscopic data, tackling a persistent issue in the field of chemistry. Detailed in a preprint on arXiv (2608.14720), MACROS mimics the expert process of iterative hypothesis testing and has been trained on 100 million simulated spectra-molecule pairs along with 1.6 million experimental ones. This system accommodates various routine spectroscopic methods, including 1D NMR, and demonstrates remarkable zero-shot generalization to a wide range of real-world samples, accurately identifying synthetic compounds, natural products, and metabolites over 500 Da. Impressively, it can recover textbook spectroscopic correlations from unassigned data and shows signs of emergent chemical intuition, such as a preference for ring-first parsing. This advancement could greatly enhance molecular discovery and analysis in areas like pharmaceuticals and natural product research.
Key facts
- MACROS is a multi-agent system for automated structure elucidation.
- Trained on 100M simulated and 1.6M experimental spectra-molecule pairs.
- Supports arbitrary combinations of routine spectroscopic techniques.
- Achieves zero-shot generalization to diverse real-world samples.
- Correctly identifies synthetic compounds, natural products, and metabolites above 500 Da with 1D NMR.
- Spontaneously recovers textbook spectroscopic correlations from unassigned data.
- Exhibits emergent chemical intuition, such as ring-first parsing preference.
- Preprint available on arXiv (2608.14720).
Entities
Institutions
- arXiv