Vib2Mol: from vibrational spectra to molecular structures-a unified deep learning framework

There will be a paradigm shift in chemical and biological research, enabled by autonomous, closed-loop, real-time self-directed decision-making experimentation. Spectrum-to-structure correlation, which is to elucidate molecular structures with spectral information, is the core step in understanding experimental results and closing the loop. Recent approaches have sought to bridge retrieval and generation to improve the accuracy of spectral annotation. However, these methods typically rely on fixed workflows that fail to accommodate the broader and more varied demands of spectrum-to-structure analysis, leaving the diverse prior knowledge provided by researchers largely underutilized. In this study, we proposed Vib2Mol, a unified deep learning framework designed to flexibly handle diverse spectrum-to-structure tasks according to the available prior knowledge. Vib2Mol not only achieves state-of-the-art performance in analyzing both theoretical and experimental infrared and Raman spectra across well-established benchmarks, but also demonstrates promising capabilities in elucidating reaction products from mixed spectra and in sequencing peptides from experimental spectra. These results position vibrational spectroscopy as a powerful guide for autonomous scientific discovery.