Elamur Research⌕
☰
home / papers / 2026 / REAP: Towards Effective Training-Free Chemical Reasoning with Explicit Atomic Priors

REAP: Towards Effective Training-Free Chemical Reasoning with Explicit Atomic Priors

Paper · 2026 · Mingxu Zhang, Dazhong Shen, Qi Zhang et al.

Summary

Frozen general-purpose LLMs often fail at precise chemical reasoning. They see molecules only as SMILES strings or coarse descriptors. Fine-tuning can add domain knowledge, but it locks that knowledge into weights, makes updates expensive, and can erode general skills.

REAP is a training-free alternative that gives the model explicit, retrievable chemical evidence. The authors build an atom-level textual knowledge base from PubChem molecules. Each atom's environment is mapped to a discrete token through a vector-quantized codebook (the AtomDisc tokenizer). Each token is linked to a readable profile of physicochemical statistics.

At inference time, REAP runs in three stages:

  • The LLM first decides which atom- and molecule-level features matter for the task.
  • Similar molecules with known labels are then retrieved.
  • The selected priors and these analogues are assembled into a structured prompt for the final prediction.

On molecular property benchmarks, REAP beats prompting baselines such as CoT and MolRAG and competes with specialized trained models. Its predictions also cite quantitative, checkable evidence. This suggests that external knowledge can stand in for parameter updates in scientific tasks.

Key points

  • Knowledge base: atoms are discretized into tokens such as "A240" by a pre-trained molecular encoder plus a codebook. Each token indexes physicochemical statistics, for example hydrogen-bond acceptor/donor ratios, aromatic ratio, ring ratio, Gasteiger charge, and mixture entropy. Molecule-level priors (TPSA, LogP, MolWt, HBA/HBD) come from RDKit.
  • Pipeline: feature selection uses three LLM rounds, but only once per task. Analogue retrieval is a CPU-only similarity search. Each sample then needs a single LLM call.
  • Cost: about 1.01–1.02 LLM calls per sample, versus 1.00 for the baselines. Total tokens are roughly 1,620, compared with about 1,164 for MolRAG; the overhead comes from longer prompts.
  • Small backbone: with Llama-2-7B, REAP reaches 0.828 AUROC on BACE and 0.783 on BBBP. MolRAG scores 0.767 and 0.629. The strongest trained model shown, LLM4SD, scores 0.838 and 0.756, so REAP is close on BACE and ahead on BBBP.
  • Limitations: only property prediction was evaluated. The faithfulness of the generated explanations is untested, and the token budget is relatively high.

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

🔬Shanghai Artificial Intelligence LaboratoryShanghai · China152 works · 13,166 citations🎓The Hong Kong University of Science and Technology (Guangzhou)Guangzhou · China67 works · 1,765 citations🎓National University of Defense TechnologyChangsha · China20 works · 1,339 citations🎓Nanjing University of Aeronautics and AstronauticsNanjing · China6 works · 2 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksComputer Vision and Pattern Recognition · 3,486 worksComputational Theory and Mathematics · 321 worksMaterials Science · 31 worksMaterials Chemistry · 31 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Mingxu Zhang, Dazhong Shen, Qi Zhang and Ying Sun. 2026. REAP: Towards Effective Training-Free Chemical Reasoning with Explicit Atomic Priors. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.97

@inproceedings{zhang2026reap,
  title = {{REAP: Towards Effective Training-Free Chemical Reasoning with Explicit Atomic Priors}},
  author = {Mingxu Zhang, Dazhong Shen, Qi Zhang and Ying Sun},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.97},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.97}
}

References

47 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  2. He Cao, Zijing Liu, Xingyu Lu, Yuan Yao, and Yu Li. 2023. Instructmol: Multi-modal integration for building a versatile and reliable molecular assistant in drug discovery. arXiv preprint arXiv:2311.16208.
  3. Zhe Chen, Zhe Fang, Wenhao Tian, Zhaoguang Long, Changzhi Sun, Yuefeng Chen, Hao Yuan, Honglin Li, and Man Lan. 2025. Reactgpt: Understanding of chemical reactions via in-context tuning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 84–92. find
  4. Shuting Cui, Ying Sun, Yuting Zhang, Qingxin Meng, and Hengshu Zhu. 2026. Llm-enhanced career knowledge graph understanding for job mobility prediction. ACM Transactions on Management Information Systems. find
  5. John S Delaney. 2004. Esol: estimating aqueous solubility directly from molecular structure. Journal of chemical information and computer sciences, 44(3):1000–1005. find
  6. Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and Jason Weston. 2024. Chain-of-verification reduces hallucination in large language models. In Findings of the association for computational linguistics: ACL 2024, pages 3563–3578.
  7. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, and 1 others. 2024. A survey on in-context learning. In Proceedings of the 2024 conference on empirical methods in natural language processing, pages 1107–1128.
  8. Zheng Gong and Ying Sun. 2025. Outlier-aware post-training quantization for discrete graph diffusion models. In Forty-second International Conference on Machine Learning. find
  9. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. 2020. Retrieval augmented language model pre-training. In International conference on machine learning, pages 3929–3938. PMLR.
  10. Yang Ji, Ying Sun, Yuting Zhang, Zhigaoyuan Wang, Yuanxin Zhuang, Zheng Gong, Dazhong Shen, Chuan Qin, Hengshu Zhu, and Hui Xiong. 2025. A comprehensive survey on self-interpretable neural networks. Proceedings of the IEEE. find
  11. Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. 2023. Active retrieval augmented generation. In Proceedings of the 2023
  12. Conference on Empirical Methods in Natural Language Processing, pages 7969–7992. Sunghwan Kim, Jie Chen, Tiejun Cheng, Asta Gindulyte, Jia He, Siqian He, Qingliang Li, Benjamin A Shoemaker, Paul A Thiessen, Bo Yu, and 1 others. 2023. find
  13. Pubchem 2023 update. Nucleic acids research, 51(D1):D1373–D1380. find
  14. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199– 22213.
  15. Seul Lee, Karsten Kreis, Srimukh Veccham, Meng Liu, Danny Reidenbach, Saee Paliwal, Arash Vahdat, and Weili Nie. 2024. Molecule generation with fragment retrieval augmentation. Advances in Neural Information Processing Systems, 37:132463–132490. find
  16. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459–…
  17. Shenghua Li and Yingwu Li. 2026. Severing the health-impairment chain: Psychological distress and cross-level safety climate associations with safety behavior. Acta Psychologica, 265:106773. find
  18. Shengchao Liu, Weili Nie, Chengpeng Wang, Jiarui Lu, Zhuoran Qiao, Ling Liu, Jian Tang, Chaowei Xiao, and Animashree Anandkumar. 2023a. Multimodal molecule structure–text model for text-based retrieval and editing. Nature Machine Intelligence, 5(12):1447–1457. find
  19. Shengchao Liu, Hanchen Wang, Weiyang Liu, Joan Lasenby, Hongyu Guo, and Jian Tang. 2022. Pre-training molecular graph representation with 3d geometry. In International Conference on Learning Representations. find
  20. Shengchao Liu, Jiongxiao Wang, Yijin Yang, Chengpeng Wang, Ling Liu, Hongyu Guo, and Chaowei Xiao. 2024. Conversational drug editing using retrieval and domain feedback. In The twelfth international conference on learning representations. find
  21. Zequn Liu, Wei Zhang, Yingce Xia, Lijun Wu, Shufang Xie, Tao Qin, Ming Zhang, and Tie-Yan Liu. 2023b. Molxpt: Wrapping molecules with text for generative pre-training. arXiv preprint arXiv:2305.10688. find
  22. Zhiyuan Liu, Sihang Li, Yanchen Luo, Hao Fei, Yixin Cao, Kenji Kawaguchi, Xiang Wang, and Tat-Seng Chua. 2023c. Molca: Molecular graph-language modeling with cross-modal projector and uni-modal adapter. arXiv preprint arXiv:2310.12798. find
  23. Jieyi Long. 2023. Large language model guided tree-of-thought. arXiv preprint arXiv:2305.08291. find
  24. Yizhen Luo, Kai Yang, Massimo Hong, Xing Yi Liu, and Zaiqing Nie. 2023. Molfm: A multimodal molecular foundation model. arXiv preprint arXiv:2307.09484. — [arXiv:2307.09484](https://arxiv.org/abs/2307.09484) find
  25. Andres M. Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D White, and Philippe Schwaller. 2024. Augmenting large language models with chemistry tools. Nature Machine Intelligence, 6(5):525– 535. find
  26. David L Mobley and J Peter Guthrie. 2014. Freesolv: a database of experimental and calculated hydration free energies, with input files. Journal of computer-aided molecular design, 28:711–720. find
  27. Kristof Schütt, Pieter-Jan Kindermans, Huziel Enoc Sauceda Felix, Stefan Chmiela, Alexandre Tkatchenko, and Klaus-Robert Müller. 2017. Schnet: A continuous-filter convolutional neural network for modeling quantum interactions. Advances in neural information processing systems, 30. find
  28. Murat Cihan Sorkun, Abhishek Khetan, and Süleyman Er. 2019. Aqsoldb, a curated reference set of aqueous solubility and 2d descriptors for a diverse set of compounds. Scientific data, 6(1):143. find
  29. Bing Su, Dazhao Du, Zhao Yang, Yujie Zhou, Jiangmeng Li, Anyi Rao, Hao Sun, Zhiwu Lu, and Ji-Rong Wen. 2022. A molecular multimodal foundation model associating molecule graphs with natural language. arXiv preprint arXiv:2209.05481. — [arXiv:2209.05481](https://arxiv.org/abs/2209.05481) find
  30. Jike Wang, Rui Qin, Mingyang Wang, Meijing Fang, Yangyang Zhang, Yuchen Zhu, Qun Su, Qiaolin Gou, Chao Shen, Odin Zhang, and 1 others. 2025. Tokenmol 1.0: tokenized drug design with large language models. Nature Communications, 16(1):4416. find
  31. Limei Wang, Kaveh Hassani, Si Zhang, Dongqi Fu, Baichuan Yuan, Weilin Cong, Zhigang Hua, Hao Wu, Ning Yao, and Bo Long. 2024. Learning graph quantized tokenizers for transformers. arXiv preprint arXiv:2410.13798. — [arXiv:2410.13798](https://arxiv.org/abs/2410.13798) find
  32. Ning-Ning Wang, Jie Dong, Yin-Hua Deng, Min-Feng Zhu, Ming Wen, Zhi-Jiang Yao, Ai-Ping Lu, Jian-Bing Wang, and Dong-Sheng Cao. 2016. Adme properties evaluation in drug discovery: prediction of caco-2 cell permeability using a combination of nsga-ii and boosting. Journal of chemical information and m… find
  33. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  34. Zhenqin Wu, Bharath Ramsundar, Evan N Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S Pappu, Karl Leswing, and Vijay Pande. 2018. Moleculenet: a benchmark for molecular machine learning. Chemical science, 9(2):513–530. find
  35. Ziting Xian, Jiawei Gu, Lingbo Li, and Shangsong Liang. 2025. Molrag: unlocking the power of large language models for molecular property prediction. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15513–15531. find
  36. Haoran Xin, Ying Sun, Chao Wang, and Hui Xiong. 2025. Llmcdsr: Enhancing cross-domain sequential recommendation with large language models. ACM Transactions on Information Systems, 43(5):1–33. find
  37. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of thoughts: Deliberate problem solving with large language models. Advances in neural information processing systems, 36:11809–11822. find
  38. Juzheng Zhang, Yatao Bian, Yongqiang Chen, and Quanming Yao. 2024. Unimot: Unified molecule-text language model with discrete token representation. arXiv preprint arXiv:2408.00863. — [arXiv:2408.00863](https://arxiv.org/abs/2408.00863) find
  39. Mingxu Zhang, Dazhong Shen, and Ying Sun. 2025a. Atomdisc: An atom-level tokenizer that boosts molecular llms and reveals structure–property associations. Preprint, arXiv:2512.03080. — [arXiv:2512.03080](https://arxiv.org/abs/2512.03080) find
  40. Mingxu Zhang, Huicheng Zhang, Jiaming Ji, Yaodong Yang, and Ying Sun. 2026. Enhance the safety in reinforcement learning by adrc lagrangian methods. arXiv preprint arXiv:2601.18142. — [arXiv:2601.18142](https://arxiv.org/abs/2601.18142) find
  41. Zimin Zhang, Qianli Wu, Botao Xia, Fang Sun, Ziniu Hu, Yizhou Sun, and Shichang Zhang. 2025b. Automated molecular concept generation and labeling with large language models. In Proceedings of the 31st International Conference on Computational Linguistics, pages 6918–6936, Abu Dhabi, UAE. Association… find
  42. Zihan Zhao, Da Ma, Lu Chen, Liangtai Sun, Zihao Li, Yi Xia, Bo Chen, Hongshen Xu, Zichen Zhu, Su Zhu, and 1 others. 2024. Chemdfm: a large language foundation model for chemistry. arXiv preprint arXiv:2401.14818. — [arXiv:2401.14818](https://arxiv.org/abs/2401.14818) find
  43. Yizhen Zheng, Huan Yee Koh, Jiaxin Ju, Anh TN Nguyen, Lauren T May, Geoffrey I Webb, and Shirui Pan. 2025. Large language models for scientific discovery in molecular property prediction. Nature Machine Intelligence, pages 1–11. find
  44. Gengmo Zhou, Zhifeng Gao, Qiankun Ding, Hang Zheng, Hongteng Xu, Zhewei Wei, Linfeng Zhang, and Guolin Ke. 2023. Uni-mol: A universal 3d molecular representation learning framework. find
  45. Hao Zhu, Todd M Martin, Lin Ye, Alexander Sedykh, Douglas M Young, and Alexander Tropsha. 2009. Quantitative structure- activity relationship modeling of rat acute toxicity by oral exposure. Chemical research in toxicology, 22(12):1913–1921. find
  46. Yuanxin Zhuang, Dazhong Shen, and Ying Sun. 2025. Moleditrl: Structure-preserving molecular editing via discrete diffusion and reinforcement learning. arXiv preprint arXiv:2505.20131. — [arXiv:2505.20131](https://arxiv.org/abs/2505.20131) find
  47. Yuanxin Zhuang, Dazhong Shen, and Ying Sun. 2026. Ngtm: Substructure-based neural graph topic model for interpretable graph generation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 29251–29259. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-reap-effective-training-free-chemical/