Breaking the Modality Barrier: Generative Modeling for Accurate Molecule Retrieval from Mass Spectra
Yiwen Zhang, Keyan Ding, Yihang Wu, Xiang Zhuang, Yi Yang, Qiang Zhang, Huajun Chen · 2025
Read Breaking the Modality Barrier: Generative Modeling for Accurate Molecule Retrieval from Mass Spectra by Yiwen Zhang, Keyan Ding, Yihang Wu, Xiang Zhuang, Yi Yang, Qiang Zhang, Huajun Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-06259.txt.
Description
Yiwen Zhang and co-authors improve how chemists identify unknown compounds from tandem mass spectra. Matching a spectrum against a library fails when the compound is missing from that library, and newer cross-modal methods often misalign spectra with structures. Their framework, GLMR, uses generative modeling to close that gap.
First, a contrastive model shortlists candidate molecules for the measured spectrum. Then a generative language model uses those candidates plus the spectrum to draft refined structures, which re-rank the shortlist by similarity. On the MassSpecGym benchmark and a new twenty-thousand-entry dataset, GLMR improved top-one accuracy by over forty percent relative to prior methods and generalized well.
Catalog record
| CREATOR | Yiwen Zhang, Keyan Ding, Yihang Wu, Xiang Zhuang, Yi Yang, Qiang Zhang, Huajun Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-06259