{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T16:08:21Z","timestamp":1790698101823,"version":"4.1.0"},"reference-count":21,"publisher":"American Chemical Society (ACS)","issue":"38","license":[{"start":{"date-parts":[[2026,9,14]],"date-time":"2026-09-14T00:00:00Z","timestamp":1789344000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000165","name":"Division of Chemistry","doi-asserted-by":"publisher","award":["1955260"],"award-info":[{"award-number":["1955260"]}],"id":[{"id":"10.13039\/100000165","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000057","name":"National Institute of General Medical Sciences","doi-asserted-by":"publisher","award":["R01GM147673"],"award-info":[{"award-number":["R01GM147673"]}],"id":[{"id":"10.13039\/100000057","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000057","name":"National Institute of General Medical Sciences","doi-asserted-by":"publisher","award":["R01GM149705"],"award-info":[{"award-number":["R01GM149705"]}],"id":[{"id":"10.13039\/100000057","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,9,29]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Optical chemical structure recognition (OCSR) is critical for converting 2D molecular diagrams from printed literature into machine-readable formats. While vision-language models have shown promise in end-to-end OCR tasks, their direct application to OCSR remains challenging, and direct full-parameter supervised fine-tuning often fails. In this work, we adapt DeepSeek-OCR-2 for molecular optical recognition by formulating the task as an image-conditioned SMILES generation. To overcome training instabilities, we propose a two-stage progressive supervised fine-tuning strategy: starting with parameter-efficient LoRA and transitioning to selective full-parameter fine-tuning with split learning rates. We train our model on a large-scale corpus combining synthetic renderings from PubChem and realistic patent images from USPTO-MOL to improve the coverage and robustness. Our fine-tuned model, MolSeek-OCR, demonstrates competitive capabilities, achieving exact-match accuracies comparable to those of the best-performing image-to-sequence model. However, it remains inferior to state-of-the-art image-to-graph models. Exploratory post-training shows mixed results: ReFT provides small benchmark-dependent gains, whereas GSPO does not yield stable improvement and can collapse under extended optimization.<\/jats:p>","DOI":"10.1021\/acsomega.6c04159","type":"journal-article","created":{"date-parts":[[2026,9,11]],"date-time":"2026-09-11T22:01:34Z","timestamp":1789164094000},"page":"57132-57138","source":"Crossref","is-referenced-by-count":0,"title":["Fine-Tuning DeepSeek-OCR-2\nfor Molecular Structure\nRecognition"],"prefix":"10.1021","volume":"11","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-5702-2847","authenticated-orcid":true,"given":"Haocheng","family":"Tang","sequence":"first","affiliation":[{"name":"University of Pittsburgh , , , ,","place":["Pittsburgh, Pennsylvania, United States, 15261"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xingyu","family":"Dang","sequence":"additional","affiliation":[{"name":"Princeton University , , , ,","place":["Princeton, New Jersey, United States, 08540"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9607-8229","authenticated-orcid":true,"given":"Junmei","family":"Wang","sequence":"additional","affiliation":[{"name":"University of Pittsburgh , , , ,","place":["Pittsburgh, Pennsylvania, United States, 15261"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"316","published-online":{"date-parts":[[2026,9,14]]},"reference":[{"key":"2026092911090851500_cit1","unstructured":"Wei, H.; Sun, Y.; Li, Y.\n          DeepSeek-OCR: Contexts Optical Compression. 2025, arXiv:2510.18234. arXiv.org e-Printarchive. https:\/\/arxiv.org\/abs\/2510.18234."},{"key":"2026092911090851500_cit2","unstructured":"Wei, H.; Sun, Y.; Li, Y.\n          DeepSeek-OCR 2: Visual Causal Flow. 2026, arXiv:2601.20552. arXiv.org e-Printarchive. https:\/\/arxiv.org\/abs\/2601.20552."},{"key":"2026092911090851500_cit3","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-40782-0","article-title":"DECIMER.ai:\nan open platform for automated optical chemical structure identification,\nsegmentation and recognition in scientific publications","volume":"14","author":"Rajan","year":"2023","journal-title":"Nat. Commun."},{"key":"2026092911090851500_cit4","doi-asserted-by":"publisher","first-page":"1925","DOI":"10.1021\/acs.jcim.2c01480","article-title":"MolScribe:\nRobust Molecular Structure Recognition with Image-to-Graph Generation","volume":"63","author":"Qian","year":"2023","journal-title":"J. Chem. Inf. Model."},{"key":"2026092911090851500_cit5","doi-asserted-by":"publisher","DOI":"10.1186\/s13321-020-00465-0","article-title":"A review of\noptical chemical structure recognition tools","volume":"12","author":"Rajan","year":"2020","journal-title":"J. Cheminf."},{"key":"2026092911090851500_cit6","doi-asserted-by":"publisher","first-page":"1017","DOI":"10.1021\/acs.jcim.8b00669","article-title":"Molecular Structure\nExtraction from Documents Using Deep Learning","volume":"59","author":"Staker","year":"2019","journal-title":"J. Chem. Inf. Model."},{"key":"2026092911090851500_cit7","doi-asserted-by":"publisher","first-page":"D1202","DOI":"10.1093\/nar\/gkv951","article-title":"PubChem Substance and Compound databases","volume":"44","author":"Kim","year":"2016","journal-title":"Nucleic Acids Res."},{"key":"2026092911090851500_cit8","doi-asserted-by":"crossref","unstructured":"Yoo, S.; Kwon, O.; Lee, H. In  Image-to-Graph Transformers for Chemical\nStructure Recognition; ICASSP 2022 - 2022 IEEE International\nConference on Acoustics, Speech and Signal Processing (ICASSP) IEEE, 2022; pp 3393\u20133397.","DOI":"10.1109\/ICASSP43922.2022.9746088"},{"key":"2026092911090851500_cit9","doi-asserted-by":"publisher","DOI":"10.1080\/27660400.2023.2260300","article-title":"Prompt engineering\nof GPT-4 for chemical research: what can\/cannot be done?","volume":"3","author":"Hatakeyama-Sato","year":"2023","journal-title":"Sci. Technol. Adv. Mater.:Methods"},{"key":"2026092911090851500_cit10","doi-asserted-by":"publisher","first-page":"23","DOI":"10.1016\/j.synbio.2024.07.004","article-title":"Integrating chemistry\nknowledge in\nlarge language models via prompt engineering","volume":"10","author":"Liu","year":"2025","journal-title":"Synth. Syst. Biotechnol."},{"key":"2026092911090851500_cit11","doi-asserted-by":"publisher","first-page":"10311","DOI":"10.1021\/acs.jcim.5c01737","article-title":"Auxiliary\nDiscriminator\nSequence Generative Adversarial Networks for Few Sample Molecule Generation","volume":"65","author":"Tang","year":"2025","journal-title":"J. Chem. Inf. Model."},{"key":"2026092911090851500_cit12","unstructured":"Zhang, D.; Liu, W.; Tan, Q.; Chen, J.; Yan, H.; Yan, Y.; Li, J.; Huang, W.; Yue, X.; Zhou, D.; Zhang, S.; Su, M.; Zhong, H.; Li, Y.; Ouyang, W.\n          ChemLLM:\nA Chemical Large Language Model. 2024, arXiv:2402.06852. arXiv.org e-Printarchive. https:\/\/arxiv.org\/abs\/2402.06852."},{"key":"2026092911090851500_cit13","doi-asserted-by":"publisher","DOI":"10.1016\/j.xcrp.2025.102523","article-title":"Developing ChemDFM as a large language foundation\nmodel for chemistry","volume":"6","author":"Zhao","year":"2025","journal-title":"Cell Rep. Phys. Sci."},{"key":"2026092911090851500_cit14","doi-asserted-by":"publisher","DOI":"10.34133\/research.0827","article-title":"BatGPT-Chem: A Foundation Large Model\nfor Chemical Engineering","volume":"8","author":"Yang","year":"2025","journal-title":"Research"},{"key":"2026092911090851500_cit15","doi-asserted-by":"publisher","first-page":"8380","DOI":"10.1039\/D4SC00966E","article-title":"nach0: multimodal natural and chemical languages foundation\nmodel","volume":"15","author":"Livne","year":"2024","journal-title":"Chem. Sci."},{"key":"2026092911090851500_cit16","unstructured":"Cao, H.; Liu, Z.; Lu, X.; Yao, Y.; Li, Y. In  InstructMol: Multi-Modal Integration for\nBuilding a Versatile\nand Reliable Molecular Assistant in Drug Discovery; Proceedings\nof the 31st International Conference on Computational Linguistics Association for Computational Linguistics, 2023."},{"key":"2026092911090851500_cit17","doi-asserted-by":"crossref","unstructured":"Jiang, L.; Sun, S.; Qi, B.; Fu, Y.; Xu, X.; Li, Y.; Zhou, D.; Fu, T.\n          Chem3DLLM:\n3D\nMultimodal Large Language Models for Chemistry. 2025, arXiv:2508.10696. arXiv.org e-Printarchive. https:\/\/arxiv.org\/abs\/2508.10696.","DOI":"10.2139\/ssrn.7135690"},{"issue":"1","key":"2026092911090851500_cit18","first-page":"415","article-title":"ChemVLM: Exploring the\nPower of Multimodal Large Language Models in Chemistry Area","volume":"39","author":"Li","year":"2025","journal-title":"Proc. AAAI Conf. Artif. Intell."},{"key":"2026092911090851500_cit19","unstructured":"Wang, J.; He, Y.; Yang, H.; Wu, J.; Ge, L.; Wei, X.; Wang, Y.; Li, L.; Ao, H.; Liu, C.; Wang, B.; Wu, L.; He, C.\n          GTR-CoT: Graph Traversal as Visual Chain of Thought for Molecular\nStructure Recognition. 2026, arXiv:2506.07553.\narXiv.org e-Printarchive. https:\/\/arxiv.org\/abs\/2506.07553."},{"key":"2026092911090851500_cit20","doi-asserted-by":"crossref","first-page":"498","DOI":"10.1007\/978-3-031-19815-1_29","article-title":"OCR-free Document Understanding Transformer","volume-title":"Computer Vision \u2013 ECCV 2022","author":"Kim","year":"2022"},{"key":"2026092911090851500_cit21","unstructured":"Blecher, L.; Cucurull, G.; Scialom, T.; Stojnic, R. In  Nougat: Neural Optical\nUnderstanding for Academic Documents; International Conference\non Learning Representations 2024 ICLR, 2023."}],"container-title":["ACS Omega"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/pubs.acs.org\/doi\/pdf\/10.1021\/acsomega.6c04159","content-type":"application\/pdf","content-version":"vor","intended-application":"unspecified"},{"URL":"https:\/\/pubs.acs.org\/acsodf\/article-pdf\/11\/38\/57132\/68489419\/acsomega.6c04159.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/pubs.acs.org\/acsodf\/article-pdf\/11\/38\/57132\/68489419\/acsomega.6c04159.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T15:50:19Z","timestamp":1790697019000},"score":1,"resource":{"primary":{"URL":"https:\/\/pubs.acs.org\/acsodf\/article\/11\/38\/57132\/5428738\/Fine-Tuning-DeepSeek-OCR-2-for-Molecular-Structure"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,14]]},"references-count":21,"journal-issue":{"issue":"38","published-online":{"date-parts":[[2026,9,14]]},"published-print":{"date-parts":[[2026,9,29]]}},"URL":"https:\/\/doi.org\/10.1021\/acsomega.6c04159","relation":{},"ISSN":["2470-1343"],"issn-type":[{"value":"2470-1343","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,14]]}}}