Reproduce “RxnMiner: A Multimodal Agent for Structured Reaction Data Extraction from Chemistry Figures” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinbo Hu, Songren Zhang, Chuanxun Fu, Jinrui Huang, Zhen Hao Wong, Dongxu Cui, Haoran Fan, Muhao Ge and 14 more
RxnMiner coordinates molecular recognition, entry-level evidence linking, and template reconstruction to extract reaction records from figures where structures and experimental details are distributed across schemes, tables, and notes. On 603 ChemRxn-Bench figures, Qwen3.8-Flash-Next with RxnMiner reaches 71.0% stereochemistry-aware macro-F1, versus 21.0% for direct prediction and 25.0% for ChemEAGLE; the score tests molecular identity and pairing, not conditions or outcomes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.