Extracting structure-activity relationships from patents with large language models.
6 affiliations
- 1State Key Laboratory of Mechanism and Quality of Chinese Medicine, University of Macau, Room 1049, E12, Macao SAR, 999078, China.
- 2Faculty of Computing, School of Computer Science and Technology (Shenzhen), Harbin Institute of Technology, Building L, HIT Campus, Shenzhen UniversityTown, Xili, Shenzhen, China.
- 3Faculty of Computing, School of Computer Science and Technology (Shenzhen), Harbin Institute of Technology, Building L, HIT Campus, Shenzhen UniversityTown, Xili, Shenzhen, China. [email protected].
- 4State Key Laboratory of Mechanism and Quality of Chinese Medicine, University of Macau, Room 1049, E12, Macao SAR, 999078, China. [email protected].
- 5Centre for Pharmaceutical Regulatory Sciences, Institute of Chinese Medical Sciences, University of Macau, Macao SAR, China. [email protected].
- 6DPM, Faculty of Health Sciences, University of Macau, Macao SAR, China. [email protected].
Abstract
Background
Persistent clinical failures in drug development arise from multiple factors, including imbalanced molecular properties, underscoring the need for systematic structure-activity relationship analysis to link structural modifications with optimized efficacy, selectivity, and safety of small-molecule drugs. Compared with academic literature, drug-related patent documents contain earlier and richer structure-activity relationship information that can support preclinical decision making. However, extracting structure-activity relationship information from patents is technically challenging due to ambiguous terminology, fragmented tables, dispersed compound descriptions, and the need to correctly link compound identities with activity endpoints. To address these issues, we developed PatSAR, a prompt-engineering-centered LLM workflow implemented through the ChatGPT web interface, in which finalized task-specific prompts guide document-by-document extraction of SAR information from patent PDF documents.Results
When evaluated on 214 PARP-related patents containing 8226 expert-annotated SAR entities, PatSAR achieved a recall of 79.05%, a precision of 99.65%, and an F1 score of 88.16%. In a multi-target generalization test set of 200 recent patents across ten non-PARP small-molecule targets, PatSAR achieved a recall of 85.25%, a precision of 98.97%, and an F1 score of 91.60%. In an external comparison on DocSAR-Patent-96, PatSAR achieved a Table Recall of 81.86%, compared with the reported Doc2SAR result of 74.09% on the patent subset.Conclusions
These results support the potential of large language models for patent information extraction and highlight the role of prompt engineering in improving task-specific extraction performance. The proposed PatSAR framework illustrates the promise of hybrid human-AI workflows for SAR curation and patent analysis.
Related
- Design, Amino-Acid-Derived Analogues of FTY-720 and OSU-2S as PP2A Activators: Structure-Activity Relationships and Identification of Lead Compound XML-3D
- Soybean paste functional peptides: Fermentation-driven formation, characterization, classification, structure-activity relationships, and multi-omics insights.
- Structure–Activity–Property Relationships of MZ1-Based Trivalent PROTACs Incorporating a 1,2,5-Trisubstituted Benzene Linker
- Extraction, Structure–Activity Relationships, and Modification of Plant-Derived Polysaccharides: An Integrated Framework for Evidence-Based Method Selection and Bioactivity Optimization—A Review
- Fragment‐Linked Purine Derivatives as Inhibitors of (p)ppGpp Synthetases: Design, Synthesis, and Structure‐Activity Relationships
BibTeX
@article{zheng2026extracting,
title = {{Extracting structure-activity relationships from patents with large language models.}},
author = {Shaolong Zheng and Gengyuan Yao and Hong Cai and Ningxuan Ma and Xiaofeng Zhang and Yuanjia Hu},
journal = {BMC bioinformatics},
year = {2026},
month = aug,
doi = {10.1186/s12859-026-06587-w},
url = {https://europepmc.org/article/MED/42958516},
note = {CC BY},
}
RIS
TY - JOUR
TI - Extracting structure-activity relationships from patents with large language models.
AU - Shaolong Zheng
AU - Gengyuan Yao
AU - Hong Cai
AU - Ningxuan Ma
AU - Xiaofeng Zhang
AU - Yuanjia Hu
PY - 2026
DA - 2026/08/03
JO - BMC bioinformatics
DO - 10.1186/s12859-026-06587-w
UR - https://europepmc.org/article/MED/42958516
AB - Background Persistent clinical failures in drug development arise from multiple factors, including imbalanced molecular properties, underscoring the need for systematic structure-activity relationship analysis to link structural modifications with optimized efficacy, selectivity, and safety of small-molecule drugs. Compared with academic literature, drug-related patent documents contain earlier and richer structure-activity relationship information that can support preclinical decision making. However, extracting structure-activity relationship information from patents is technically challenging due to ambiguous terminology, fragmented tables, dispersed compound descriptions, and the need to correctly link compound identities with activity endpoints. To address these issues, we developed PatSAR, a prompt-engineering-centered LLM workflow implemented through the ChatGPT web interface, in which finalized task-specific prompts guide document-by-document extraction of SAR information from patent PDF documents.Results When evaluated on 214 PARP-related patents containing 8226 expert-annotated SAR entities, PatSAR achieved a recall of 79.05%, a precision of 99.65%, and an F1 score of 88.16%. In a multi-target generalization test set of 200 recent patents across ten non-PARP small-molecule targets, PatSAR achieved a recall of 85.25%, a precision of 98.97%, and an F1 score of 91.60%. In an external comparison on DocSAR-Patent-96, PatSAR achieved a Table Recall of 81.86%, compared with the reported Doc2SAR result of 74.09% on the patent subset.Conclusions These results support the potential of large language models for patent information extraction and highlight the role of prompt engineering in improving task-specific extraction performance. The proposed PatSAR framework illustrates the promise of hybrid human-AI workflows for SAR curation and patent analysis.
SN - 1471-2105
ER -