A Pre-trained Reaction Embedding Descriptor Capturing Bond Transformation Patterns
Weiqi Liu, Fenglei Cao, Yuan Qi, Li-Cheng Xu · 2026
Read A Pre-trained Reaction Embedding Descriptor Capturing Bond Transformation Patterns by Weiqi Liu, Fenglei Cao, Yuan Qi, Li-Cheng Xu (2026) 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-2601-03689.txt.
Description
Weiqi Liu and three co-authors introduce RXNEmb, a learned fingerprint for whole chemical reactions. Data-driven reaction prediction needs descriptors that capture what a reaction does, yet general reaction-level descriptors are scarce. RXNEmb comes from a graph model pretrained to tell real reactions from fake ones with wrong bond changes, forcing it to learn how bonds form and break.
The team reclustered the standard USPTO-50k reaction dataset with the new descriptor and got groupings that follow bond-change similarity more closely than rule-based categories. Combined with dimensionality reduction, the descriptor also maps the diversity of reaction space visually. Attention analysis showed the model focusing on chemically critical sites, offering mechanistic clues. A shared reaction fingerprint could help chemists search, compare, and organize reaction data at scale.
Catalog record
| CREATOR | Weiqi Liu, Fenglei Cao, Yuan Qi, Li-Cheng Xu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03689