DianShi-RxnDB: A Large-Scale, Fine-Grained Organic Reaction Data Platform Built via a Fully Automated Pipeline for Researchers and AI Agents
Yubin Wang, Xingjian Wei, Jiang Wu, Yinfan Wang, Boyu Zhu, Lin Zhang, Jianing Yu, Huazheng Zeng and 28 others
Shanghai Artificial Intelligence Laboratory · Fudan University · Shanghai Jiao Tong University · East China Normal University · East China University of Science and Technology
Published September 6, 2026 · on the feed September 10, 2026 · ▲ 12 · arXiv 2609.06703
The abstract
High-quality structured organic reaction data are essential for developing artificial intelligence for chemistry (AI4Chem), yet much of this knowledge remains dispersed across patent text, images, and reaction schemes. We present DianShi-RxnDB, a large-scale, fine-grained organic reaction data platform built via a fully automated extraction and normalization pipeline integrating patent text, images, and reaction schemes. Its corpus covers organic synthesis patents from the USPTO and EPO published between 1976 and 2025, yielding approximately 24 million reaction instances, of which approximately 14.8 million (61.7%) pass automated qualification checks. Each instance represents a specific single-step experiment recording participants, roles, quantities, temperatures, reaction times, yields, experimental procedures, and provenance links to source patents. In a manual evaluation of 1,300 sampled qualified instances, the micro-averaged field-level accuracy was 92.95%. A matched comparison with Pistachio further indicated advantages in deduplicated record counts, representation granularity, and field-level exact agreement. The platform provides a Web research workbench for searching, filtering, comparing, and source-verifying records, and a Model Context Protocol (MCP) service offering AI agents composable structured retrieval tools. DianShi-RxnDB is available at https://dianshi.opendatalab.org.cn/ .
The Brief
2:30A short podcast summary of the paper. Written and voiced by AI.
Transcript
DianShi-RxnDB. A Large-Scale, Fine-Grained Organic Reaction Data Platform Built via a Fully Automated Pipeline for Researchers and AI Agents. By Yubin Wang and thirty-five others, from the Shanghai Artificial Intelligence Laboratory, Fudan University, and three other institutions. Published on the sixth of September. No code repository is linked to it.
DianShi-RxnDB is a large-scale, fine-grained platform of organic reaction data. It exists because high-quality, structured reaction data are essential for building artificial intelligence for chemistry, AI4Chem, and yet much of that knowledge is scattered across patent text, images and reaction schemes, in forms a machine cannot readily use.
The corpus holds approximately twenty-four million reaction instances. Of those, approximately fourteen point eight million, sixty-one point seven percent, pass automated qualification checks. Each instance is one specific single-step experiment. It records the participants and their roles, quantities, temperatures, reaction times, yields, the experimental procedure, and links back to the source patent for provenance.
The data can be reached two ways. A Web research workbench supports searching, filtering, comparing and verifying records against their sources. And a Model Context Protocol service, MCP, gives AI agents structured retrieval tools that can be composed together.
Everything comes from organic synthesis patents filed with the USPTO and the EPO, spanning nearly five decades of publication. A fully automated pipeline does the extraction and normalization, integrating the patent text, the images and the reaction schemes into one record.
Quality was checked by hand. In a manual evaluation of one thousand three hundred sampled qualified instances, micro-averaged accuracy at the field level was ninety-two point nine five percent. A matched comparison with Pistachio further indicated advantages in deduplicated record counts, in the granularity of the representation, and in exact agreement at the field level.
The title names two audiences, researchers and AI agents, and the platform serves each through its own door. Researchers get the workbench, with the ability to trace any record back to the patent it came from. Agents get the MCP service, so a model can pull structured reaction records as a tool call rather than reading patents itself.
The contribution is the conversion. Reaction knowledge that sat in patent prose and drawings now sits in structured, checkable records, tens of millions of them, each tied to its source.
Paper Quest · 7 chapters · about 40 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Decades of chemistry sit locked in patents. Your first wall: getting them out.
You want to build chemistry AI that suggests reaction conditions. The knowledge exists: decades of organic synthesis patents describe real experiments in detail.
The road ahead
- 1 · The problem ~6 min Patents hold decades of real chemistry experiments, but each one is scattered across text, images and schemes that a machine cannot search.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min An automated pipeline reads patent text, images and schemes together, turning each experiment into one structured, checked and traceable entry.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min Gather patents, extract experiments from all their parts, record each as a structured instance, check it automatically, then serve it two ways.
- 6 · The experiment ~6 min You’ll guess: When people manually checked a sample of the automatically extracted qualified reactions, what micro-averaged field-level accuracy did they find?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.