Elamur Research⌕
☰
home / papers / 2026 / BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning

BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning

Paper · 2026 · Aditya Hemant Shahane, Anuj Kumar Sirohi, Devansh Arora et al.

Summary

BiMol-Diff tackles two tasks that are usually handled by separate models: generating molecules from text descriptions and writing captions for molecules. The authors argue that autoregressive models such as T5 or GPT handle the long-range constraints of SMILES poorly, for example ring closures and branches, because early errors propagate. Standard diffusion models have a different weakness: they corrupt every position equally. BiMol-Diff is a single conditional diffusion framework that covers both directions. On the molecule side, canonical SMILES is converted into Atoms-in-SMILES (AIS) tokens and then into a serialized graph of atom–bond–atom triplets. The core idea is a token-aware noise schedule. Each token's training loss serves as a measure of how hard it is to recover, and harder tokens are noised more gently. On ChEBI-20 and M3-20M, the model improves Exact Match for molecule generation by 15.4% relative to baselines. For captioning, it achieves the top BLEU and BERTScore among compared systems. The work suggests that difficulty-aware corruption is a practical way to improve fidelity in modeling the link between molecules and language.

Key points

  • Molecules are serialized as edge lists using the special tokens [HEAD], [REL], [TAIL] and [SEP]. Predicted graphs are mapped back deterministically to canonical SMILES for evaluation.
  • In the M3-20M captioning ablation, the token-aware linear schedule beats a uniform sqrt schedule:
    • BLEU rises from 0.495 to 0.567.
    • ChrF++ rises from 0.682 to 0.734.
    • METEOR rises from 0.531 to 0.626.
    • Linear mapping outperforms cosine mapping.
  • Among tokenizers, AIS scores best, atom-level tokenization comes second, and regex-based SMILES tokenization performs worst.
  • Inference costs:
    • The 63M-parameter model with T=2000 steps needs 89 s per batch of 50 on one V100.
    • This is 3.6× faster than DiffuSeq (91M parameters, 317 s), with higher BLEU.
    • It also beats Text+Chem T5 on BLEU (0.567 vs 0.542).
  • Quality degrades sharply with fewer than about 1000 steps. At 500 steps BLEU falls to 0.365, below MolT5-Base at 0.452.
  • The authors list two limitations: the model has no explicit graph inductive bias, and inference is slower than with autoregressive models.

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

🎓Indian Institute of Science BangaloreBengaluru · India10 works · 1,142 citations🎓Indian Institute of Technology DelhiNew Delhi · India22 works · 1,036 citations

Also on the byline: Indian Institute of Technology Ropar.

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksComputational Theory and Mathematics · 252 works

Cite

Cite as Aditya Hemant Shahane, Anuj Kumar Sirohi, Devansh Arora, Nitin Kumar, Prathosh AP and Sandeep Kumar. 2026. BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1059

@inproceedings{shahane2026bimol,
  title = {{BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning}},
  author = {Aditya Hemant Shahane, Anuj Kumar Sirohi, Devansh Arora, Nitin Kumar, Prathosh AP and Sandeep Kumar},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1059},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1059}
}

References

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

  1. deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, Minneapolis, Minnesota. Association for Co… find
  2. Marianne Arriola, Subham Sekhar Sahoo, Aaron Gokaslan, Zhihan Yang, Zhixuan Qi, Jiaqi Han, Justin T Chiu, and Volodymyr Kuleshov. 2025. Block diffusion: Interpolating between autoregressive and diffusion language models. In The Thirteenth International Conference on Learning Representations. find
  3. Viraj Bagal, Rishal Aggarwal, P. K. Vinod, and U. Deva Priyakumar. 2022. Molgpt: Molecular generation using a transformer-decoder model. Journal of Chemical Information and Modeling, 62(9):2064–2076. PMID: 34694798. find
  4. Satanjeev Banerjee and Alon Lavie. 2005. METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72, Ann Arbor, Michigan. Ass…
  5. Tian Bian, Yifan Niu, Heng Chang, Divin Yan, Junzhou Huang, Yu Rong, Tingyang Xu, Jia Li, and Hong Cheng. 2024. Hierarchical graph latent diffusion model for conditional molecule generation. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ’24, pa… find
  6. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, and… find
  7. Jinho Chang and Jong Chul Ye. 2025. LDMol: A text-to-molecule diffusion model with structurally informative latent space surpasses AR models. In Forty-second International Conference on Machine Learning. find
  8. Dimitrios Christofidellis, Giorgio Giannone, Jannis Born, Ole Winther, Teodoro Laino, and Matteo Manica. 2023. Unifying molecular and textual representations via multi-task language modelling. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machi… find
  9. Yifan Deng, Spencer S. Ericksen, and Anthony Gitter. 2025. Chemical language model linker: Blending text and molecules with modular adapters. Journal of Chemical Information and Modeling, 65(17):8944–8956. find
  10. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Carl Edwards, Tuan Lai, Kevin Ros, Garrett Honke, Kyunghyun Cho, and Heng Ji. 2022. Translation between molecules and natural language. In Proceedings of the 2022 Conference on Empirical Methods in Natural…
  11. Yin Fang, Ningyu Zhang, Zhuo Chen, Lingbing Guo, Xiaohui Fan, and Huajun Chen. 2024. Domain-agnostic molecular generation with chemical feedback. In The Twelfth International Conference on Learning Representations. find
  12. Bahare Fatemi, Jonathan Halcrow, and Bryan Perozzi. 2024. Talk like a graph: Encoding graphs for large language models. In The Twelfth International Conference on Learning Representations. find
  13. Haisong Gong, Qiang Liu, Shu Wu, and Liang Wang. 2024. Text-guided molecule generation with diffusion language model. In Proceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Sympos… find
  14. Shansan Gong, Mukai Li, Jiangtao Feng, Zhiyong Wu, and Lingpeng Kong. 2023. DiffuSeq: Sequence to sequence text generation with diffusion models. In The Eleventh International Conference on Learning Representations. find
  15. Shuhan Guo, Yatao Bian, Ruibing Wang, Nan Yin, Zhen Wang, and Quanming Yao. 2025a. Unified molecule-text language model with discrete token representation. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, pages 9205–9213. International Joint Co… find
  16. Siyuan Guo, Lexuan Wang, Chang Jin, Jinxian Wang, Han Peng, Huayang Shi, Wengen Li, Jihong Guan, and Shuigeng Zhou. 2025b. M 3 -20m: A large-scale multi-modal molecule dataset for ai-driven drug design and discovery. Journal of Bioinformatics and Computational Biology, 23(02):2550006. PMID: 40494666… find
  17. Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, pages 6840–6851. Curran Associates, Inc. find
  18. Ross Irwin, Spyridon Dimitriadis, Jiazhen He, and Esben Jannik Bjerrum. 2022. Chemformer: a pre-trained transformer for computational chemistry. Machine Learning: Science and Technology, 3(1):015022. find
  19. Sangyeup Kim, Nayeon Kim, Yinhua Piao, and Sun Kim. 2025a. GraphT5: Unified molecular graph-language modeling via multi-modal cross-token attention. Preprint, arXiv:2503.07655. — [arXiv:2503.07655](https://arxiv.org/abs/2503.07655) find
  20. Seojin Kim, Hyeontae Song, Jaehyun Nam, and Jinwoo Shin. 2025b. Training text-to-molecule models with context-aware tokenization. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 22442–22460, Suzhou, China. Association for Computational Linguistics. find
  21. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the…
  22. Xiang Lisa Li, John Thickstun, Ishaan Gulrajani, Percy Liang, and Tatsunori B. Hashimoto. 2022. Diffusion-lm improves controllable text generation. In Proceedings of the 36th International Conference on Neural Information Processing Systems, NIPS ’22, Red Hook, NY, USA. Curran Associates Inc.
  23. Gang Liu, Jiaxin Xu, Tengfei Luo, and Meng Jiang. 2024a. Graph diffusion transformers for multi-conditional molecular generation. In Advances in Neural Information Processing Systems. Curran Associates, Inc. find
  24. Pengfei Liu, Yiming Ren, Jun Tao, and Zhixiang Ren. 2024b. GIT-Mol: A multi-modal large language model for molecular science with graph, image, and text. Computers in Biology and Medicine, 171:108073. find
  25. Zequn Liu, Wei Zhang, Yingce Xia, Lijun Wu, Shufang Xie, Tao Qin, Ming Zhang, and Tie-Yan Liu. 2023a. MolXPT: Wrapping molecules with text for generative pre-training. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 1606–1616… find
  26. Zhiyuan Liu, Sihang Li, Yanchen Luo, Hao Fei, Yixin Cao, Kenji Kawaguchi, Xiang Wang, and Tat-Seng Chua. 2023b. MolCA: Molecular graph-language modeling with cross-modal projector and uni-modal adapter. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages… find
  27. Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. 2025. Large language diffusion models. Preprint, arXiv:2502.09992.
  28. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, PA. Association for Computational Linguistics.
  29. Krishna Pillutla, Lang Liu, John Thickstun, Sean Welleck, Swabha Swayamdipta, Rowan Zellers, Sewoong Oh, Yejin Choi, and Zaid Harchaoui. 2023. Mauve scores for generative models: Theory and practice. Preprint, arXiv:2212.14578. — [arXiv:2212.14578](https://arxiv.org/abs/2212.14578) find
  30. Maja Popović. 2015. chrF: character n-gram F-score for automatic MT evaluation. In Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal. Association for Computational Linguistics.
  31. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67.
  32. Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. 2015. Deep unsupervised learning using nonequilibrium thermodynamics. In Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pages 2256– 2265, Lille, F…
  33. Duong Thanh Tran, Nhat Truong Pham, Nguyen Doan Hieu Nguyen, and Balachandran Manavalan. 2024. Mol2Lang-VLM: Vision- and text-guided generative pre-trained language models for advancing molecule captioning through multimodal fusion. In Proceedings of the 1st Workshop on Language + Molecules (L+M 202… find
  34. Umit Ucak, Islambek Ashyrmamatov, and Juyong Lee. 2023. Improving the quality of chemical language model outcomes with atom-in-smiles tokenization. Journal of Cheminformatics, 15. find
  35. Clement Vignac, Igor Krawczuk, Antoine Siraudin, Bohan Wang, Volkan Cevher, and Pascal Frossard. 2023. DiGress: Discrete denoising diffusion for graph generation. In The Eleventh International Conference on Learning Representations. find
  36. David Weininger. 1988. Smiles, a chemical language and information system. 1. introduction to methodology and encoding rules. Journal of Chemical Information and Computer Sciences, 28(1):31–36. find
  37. David Weininger, Arthur Weininger, and Joseph L. Weininger. 1989. Smiles. 2. algorithm for generation of unique smiles notation. Journal of Chemical Information and Computer Sciences, 29(2):97–101. find
  38. Yuran Xiang, Haiteng Zhao, Chang Ma, and Zhi-Hong Deng. 2025. Instruction-based molecular graph generation with unified text-graph diffusion model. Preprint, arXiv:2408.09896. — [arXiv:2408.09896](https://arxiv.org/abs/2408.09896) find
  39. Zheni Zeng, Bangchen Yin, Shipeng Wang, Jiarui Liu, Cheng Yang, Haishen Yao, Xingzhi Sun, Maosong Sun, Guotong Xie, and Zhiyuan Liu. 2024. Chat-Mol: interactive molecular discovery with natural language. Bioinformatics, 40(9):btae534. find
  40. Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating text generation with bert. In International Conference on Learning Representations. find
  41. Xiaochen Zhang, Shuangxi Wang, Ying Fang, and Qiankun Zhang. 2025. MG-DIFF: A novel molecular graph diffusion model for molecular generation and optimization. PLOS ONE, 20(10):e0331450. find
  42. Zaixiang Zheng, Yifan Deng, Dongyu Xue, Yi Zhou, Fei Ye, and Quanquan Gu. 2023. Structure-informed language models are protein designers. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 42317–42338. PMLR. find
  43. Huaisheng Zhu, Teng Xiao, and Vasant G Honavar. 2024. 3M-Diffusion: Latent multi-modal diffusion for language-guided molecular structure generation. In First Conference on Language Modeling. find
  44. Rustam Zhumagambetov, Ferdinand Molnar, Vsevolod A. Peshkov, and Siamac Fazli. 2021. Transmol: repurposing a language model for molecular generation. RSC Adv., 11:25921–25932. find
  45. A mechanism, a feature notably absent in prior diffusion baselines—which adapts the noising schedule to prioritize chemically significant tokens, similar to the strategy employed by the AR-based ChemT5. A.2 Derivation find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shahane-2026-bimol-diff-unified-diffusion-framework/