Elamur Research⌕
☰
home / papers / 2026 / CrossDDI: Cross-Source Evidence-Grounded Drug-Drug Interaction Verification

CrossDDI: Cross-Source Evidence-Grounded Drug-Drug Interaction Verification

Paper · 2026 · Hui Wang, Bohao Chu and Norbert Fuhr

Summary

The paper tackles drug–drug interaction (DDI) assessment with LLMs, where answers are hard to audit and RAG outputs may not follow from the retrieved passages. The authors treat DDI assessment as claim verification over normalized drug pairs and build CrossDDI. The system first resolves drug names against DrugBank synonyms and forms an interaction claim. It then gathers stances from two paths: Path A is an exact pair lookup in DrugBank, and Path B uses an LLM to select and label sentences from PubMed articles. A fixed rule, not the LLM, issues the final verdict, so every positive prediction rests on explicit evidence. Evaluation uses 1,000 DDInter 2.0 pairs under a positive–unlabeled setup, with three LLM backbones. CrossDDI reaches recall similar to or above RAG and behaves more consistently across backbones. However, its performance depends heavily on DrugBank coverage, and the literature path is sparse and less reliable. The work matters for building traceable, auditable medication-safety tools.

Key points

  • Evaluation set: 1,000 DDInter 2.0 pairs (797 confirmed interactions, 203 unlabeled). Precision and FPR cannot be identified, so recall is the main metric and a heuristic pseudo-negative FPR is added.
  • Backbones are Qwen2.5-7B, GPT-4o and Qwen3-235B-A22B. CrossDDI recall is 0.576–0.593, and its cross-backbone recall spread is 0.018, compared with 0.066 for RAG.
  • Recall is 85.81% on pairs that DrugBank covers directly and only 18.98% on pairs it does not cover. PubMed retrieval reaches just 40.5% of confirmed positives.
  • In a single-annotator manual check of 100 Interaction predictions, 84.0% had correct stances. Cases triggered only by Path B were correct just 44.0% of the time.
  • The support-favoring arbitration rule produces a higher pseudo-negative FPR than RAG on all suites. This reflects a trade-off between recall and false positives.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of Duisburg-EssenEssen · Germany3 works · 48 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputational Theory and Mathematics · 252 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

Cite

Cite as Hui Wang, Bohao Chu and Norbert Fuhr. 2026. CrossDDI: Cross-Source Evidence-Grounded Drug-Drug Interaction Verification. BioNLP 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bionlp-1.73

@inproceedings{wang2026crossddi,
  title = {{CrossDDI: Cross-Source Evidence-Grounded Drug-Drug Interaction Verification}},
  author = {Hui Wang, Bohao Chu and Norbert Fuhr},
  booktitle = {BioNLP 2026},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-1.73},
  url = {https://doi.org/10.18653/v1/2026.bionlp-1.73}
}

References

24 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-rag: Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations (ICLR).
  2. Xin Bi, He Ma, Jianhua Li, Yuliang Ma, and Deyang Chen. 2018. A positive and unlabeled learning framework based on extreme learning machine for drug-drug interactions discovery. Journal of Ambient Intelligence and Humanized Computing. find
  3. Haohuai He, Guanxing Chen, and Calvin Yu-Chian Chen. 2022. 3dgt-ddi: 3d graph and text based neural network for drug–drug interaction prediction. Briefings in Bioinformatics, 23(3):bbac134. find
  4. Wenzhun Huang, Xiao Wang, Yunhao Chen, Changqing Yu, and Shanwen Zhang. 2025. Advancing drug-drug interactions research: integrating ai-powered prediction, vulnerable populations, and regulatory insights. Frontiers in Pharmacology, 16:1618701. find
  5. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  6. Craig Knox, Michael Wilson, Christian M. Klinger, et al. 2024. Drugbank 6.0: the drugbank knowledgebase for 2024. Nucleic Acids Research, 52(D1):D1265– D1275. find
  7. Prajwol Lamichhane, Indika Kahanda, Xudong Liu, Karthikeyan Umapathy, Sandeep Reddivari, Catherine Christie, Andrea Arikawa, and Jenifer Ross. 2023. Poster: Biofactcheck: Exploring the feasibility of explainable automated inconsistency detection in biomedical and health literature. In Proceedings of… find
  8. Dongxu Li, Yue Yang, Ziwen Cui, Hengchuang Yin, Pengwei Hu, and Lun Hu. 2026. Llm-ddi: Leveraging large language models for drug-drug interaction prediction on biomedical knowledge graph. IEEE Journal of Biomedical and Health Informatics, 30(1):773–781. find
  9. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng… find
  10. David Wadden, Kyle Lo, Lucy Lu Wang, Arman Cohan, Iz Beltagy, and Hannaneh Hajishirzi. 2022b. MultiVerS: Improving scientific claim verification with weak supervision and full-document context. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 61–76, Seattle, United Sta… find
  11. GuiShen Wang, Hui Feng, and Chen Cao. 2025a. Birnn-ddi: A drug-drug interaction event type prediction model based on bidirectional recurrent neural network and graph2seq representation. Journal of Computational Biology, 32(2):198–211. find
  12. Sheng Wang, Fangyuan Zhao, Dechao Bu, Yunwei Lu, Ming Gong, Hongjie Liu, Zhaohui Yang, Xiaoxi Zeng, Zhiyuan Yuan, Baoping Wan, et al. 2025b. Lins: A general medical q&a framework for enhancing the quality and credibility of llm-generated responses. Nature communications, 16(1):9076. find
  13. Christopher A White, Yehuda A Masturov, Eric Haunschild, Evan Michaelson, Dave R Shukla, and Paul J Cagle. 2025. Can chatgpt reliably answer the most common patient questions regarding total shoulder arthroplasty? Journal of Shoulder and Elbow Surgery, 34(5):e254–e264. find
  14. Justine Sicard, François Montastruc, Coline Achalme, Annie Pierre Jonville-Bera, Paul Songue, Marina Babin, Thomas Soeiro, Pauline Schiro, Claire de Canecaude, and Romain Barus. 2025. Can large language models detect drug–drug interactions leading to adverse drug reactions? Therapeutic Advances in D… find
  15. Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al. 2023. Large language models encode clinical knowledge. Nature, 620(7972):172–180.
  16. Yao Tian, Jiacai Yi, Ningning Wang, Chengkun Wu, Jinfu Peng, Shao Liu, Guoping Yang, and Dongsheng Cao. 2025. Ddinter 2.0: an enhanced drug interaction resource with expanded data coverage, new interaction types, and improved user interface. Nucleic Acids Research, 53(D1):D1356–D1362. find
  17. Juraj Vladika, Phillip Schneider, and Florian Matthes. 2024. Healthfc: Verifying health claims with evidence-based medical fact-checking. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 8095– 8107. find
  18. David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Iz Beltagy, Lucy Lu Wang, and Hannaneh Hajishirzi. 2022a. Scifact-open: Towards open-domain scientific claim verification. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 4719–4734.
  19. Di Wu, Jia-Chen Gu, Fan Yin, Nanyun Peng, and Kai-Wei Chang. 2024. Synchronous faithfulness monitoring for trustworthy retrieval-augmented generation. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 9390–9406, Miami, Florida, USA. Association for Comp… find
  20. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jin… — Qwen3 Technical Report
  21. Cyril Zakka, Rohan Shad, Akash Chaurasia, Alex R Dalal, Jennifer L Kim, Michael Moor, Robyn Fong, Curran Phillips, Kevin Alexander, Euan Ashley, et al. 2024. Almanac—retrieval-augmented language models for clinical medicine. Nejm ai, 1(2):AIoa2300068. find
  22. Chengcheng Zhang, Yao Lu, and Tianyi Zang. 2022. Cnn-ddi: a learning-based method for predicting drug–drug interactions using convolutional neural networks. BMC Bioinformatics, 23(Suppl 1):88. find
  23. Changpeng Zhao, Dongfang Han, Zicheng Zuo, and Turdi Tohti. 2025. Kgdb-ddi: Knowledge graph-based drug background data fusion model for drug– drug interaction prediction. Artificial Intelligence in Medicine, 168:103225. find
  24. Jing Zhu, Chao Che, Hao Jiang, Jian Xu, Jiajun Yin, and Zhaoqian Zhong. 2024. Ssf-ddi: a deep learning method utilizing drug sequence and substructure features for drug–drug interaction prediction. BMC bioinformatics, 25(1):39. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-crossddi-cross-source-evidence-grounded/