NMRAgent Enhances Molecular Structure Elucidation
- •Researchers developed NMRAgent to improve molecular structure identification using NMR spectroscopy and LLM-based reasoning.
- •The system achieved a 46.5% increase in top-1 accuracy and 0.502 higher Tanimoto similarity over existing methods.
- •NMRAgent verified its efficacy by identifying new natural products and correcting established structural misassignments.
Researchers introduced NMRAgent, an AI-powered agent designed to interpret Nuclear Magnetic Resonance (NMR) spectra for molecular structure elucidation. Unlike traditional methods that rely on database retrieval or black-box models, this system integrates LLMs with chemical knowledge graphs to provide atom-level interpretability. The agent mimics human deductive reasoning by planning its analysis, proposing structural candidates, and verifying peak-atom consistency.
Performance evaluations on a scaffold-split benchmark show that NMRAgent improves top-1 accuracy by 46.5% and Tanimoto similarity by 0.502 compared to current state-of-the-art systems. Practical utility was demonstrated by successfully identifying unknown natural products from Hydrangea davidii and Vitex trifolia, as well as correcting previous structural misassignments in scientific literature.