In this work, we introduce NMR-Solver, an automated and interpretable framework for molecular structure elucidation from 1H and 13C NMR spectra. By integrating large-scale spectral matching with physics-guided molecular optimization, the method explicitly leverages atomic-level structure–spectrum relationships in NMR. Evaluations on simulated benchmarks, curated literature datasets, and real experimental cases demonstrate strong generalization, robustness, and practical applicability. By unifying computational NMR analysis, deep learning, and chemically interpretable reasoning, NMR-Solver provides a scalable solution for automated structure determination and establishes a general paradigm for inverse problems in molecular science.
