HSPAG: A Unified and Data-efficient Molecular Property Modeling Framework via Multi-Level Contrastive Learning
Ziyu Fan, Zhijian Huang, Yahan Li, Siyuan Shen, Yunliang Wang, Zeyu Zhong, Shuhong Liu, Shuning Yang, Shangqian Wu, Min Wu, Lei Deng · 2025
Read HSPAG: A Unified and Data-efficient Molecular Property Modeling Framework via Multi-Level Contrastive Learning by Ziyu Fan, Zhijian Huang, Yahan Li, Siyuan Shen, Yunliang Wang, Zeyu Zhong, Shuhong Liu, Shuning Yang, Shangqian Wu, Min Wu, Lei Deng (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-2511-08080.txt.
Description
Ziyu Fan and co-authors propose HSPAG, a data efficient method for generating and editing molecules under property limits. Drug discovery needs molecules that meet several property goals at once, but structure property links are tangled and annotations are sparse. Note the arXiv title differs from the older evidence title for this record.
HSPAG aligns molecule structure and properties at atom, substructure, and whole molecule levels, treating SMILES strings and properties as paired views. Scaffold clustering picks representative training samples, while a variational model finds hard cases, cutting data needs. Masking based on property relevance and varied perturbations improve generation, and two real cases test editing.
Catalog record
| CREATOR | Ziyu Fan, Zhijian Huang, Yahan Li, Siyuan Shen, Yunliang Wang, Zeyu Zhong, Shuhong Liu, Shuning Yang, Shangqian Wu, Min Wu, Lei Deng |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-08080