MinerU-Chem Reports 93.02% Exact Molecular Recognition
A chemistry-focused extension to MinerU converts document regions into molecule and reaction records, outperforming the evaluated GPT-5.6-Sol comparison on a SMILES benchmark subset.
Underlying Paper
MinerU.Chem: A High-Precision System for Optical Chemical Structure and Reaction Recognition
In organic chemistry papers and patents, molecular structures, reaction schemes, and experimental conditions are often presented as molecular structure depictions, reaction diagrams, and complex tables or figures. Such information is difficult for general-purpose document parsing systems to directly convert into machine-readable data. This limits data production for organic chemistry knowledge base construction and for AI for Chemistry tasks such as reaction prediction, retrosynthesis, condition recommendation, molecular property prediction, and drug molecule design. This report introduces MinerU-Chem, a document parsing system for organic chemistry literature integrated into the MinerU online platform. Built on top of MinerU's general document parsing pipeline, MinerU-Chem adds five chemistry-specific modules: chemistry relevance filtering, molecular structure detection, molecule identifier extraction, molecular structure recognition, and reaction scheme parsing. Together, these modules convert organic-chemistry-related image regions in documents into a Molecule Summary List and a Reaction Summary List. For molecular structure recognition, MinerU-Chem uses CARBON (Complex Atomic Representation and Bonding Object Notation) as its core representation. CARBON enables recognition results to preserve both the visual layout of the original image and complex chemical semantics, while supporting the export of standard downstream formats such as MolFile and SMILES. On the SMILES-evaluable subset of MolRecBench-Wild (N=2,392), MinerU-Chem's molecular structure recognition module achieves a SMILES exact-match accuracy of 93.02%, outperforming the best evaluated comparison system, GPT-5.6-Sol (74.87%), by 18.15 percentage points. The system has been integrated into the MinerU online platform and is available at https://mineru.net/OpenSourceTools/Extractor .
Organic chemistry papers and patents often encode important information in molecular drawings, reaction schemes, and complex figures rather than in searchable text. Converting those visual elements into machine-readable data can support chemistry knowledge bases and AI applications such as reaction prediction, retrosynthesis, condition recommendation, molecular property prediction, and drug design.
Core Contribution
MinerU-Chem is a chemistry-focused document-parsing system integrated into the MinerU platform. It adds five modules to MinerU's general document pipeline: chemistry relevance filtering, molecular structure detection, molecule identifier extraction, molecular structure recognition, and reaction scheme parsing.
Together, these components convert chemistry-related image regions into a Molecule Summary List and a Reaction Summary List. Figure 2 shows examples of these two output formats.
Technical Approach
For molecular structure recognition, MinerU-Chem uses CARBON, or Complex Atomic Representation and Bonding Object Notation. According to the paper, CARBON is intended to preserve both a chemical drawing's visual layout and its chemical semantics while supporting export to standard downstream formats including MolFile and SMILES.
The system's broader design combines structure recognition with extraction of molecule identifiers and reaction schemes. This positions the pipeline as a way to turn chemistry-related document imagery into structured outputs rather than only recovering isolated molecular strings.
Results and Analysis
On the SMILES-evaluable subset of MolRecBench-Wild, containing 2,392 examples, MinerU-Chem reports 93.02% SMILES exact-match accuracy. The paper reports 74.87% for GPT-5.6-Sol, the best evaluated comparison system named in the abstract, a difference of 18.15 percentage points.
This result provides direct quantitative evidence for the molecular structure recognition component on the evaluated subset. The abstract does not provide corresponding benchmark results for every other pipeline component, such as reaction scheme parsing or molecule identifier extraction.
Scope
The reported result is specific to the SMILES-evaluable MolRecBench-Wild subset. MinerU-Chem's stated goal is broader document-to-structure conversion, but the supplied benchmark evidence is concentrated on molecular structure recognition.
The system is integrated into the MinerU online platform, providing access to its chemistry-document extraction workflow.
Evidence Box
moderateKey Claims
- •MinerU-Chem adds chemistry-specific parsing modules to the MinerU document pipeline
- •The system produces Molecule Summary Lists and Reaction Summary Lists from chemistry-related document regions
- •CARBON supports structure recognition with export to MolFile and SMILES
Key Results
- •93.02% SMILES exact-match accuracy on the 2,392-example MolRecBench-Wild evaluable subset
- •74.87% SMILES exact-match accuracy for GPT-5.6-Sol in the reported comparison
- •18.15-percentage-point lead over GPT-5.6-Sol
Limitations & Caveats
- •The reported quantitative result applies to the SMILES-evaluable MolRecBench-Wild subset
- •The supplied abstract does not report comparable benchmark metrics for all document-parsing modules
- •Broader document extraction performance should not be inferred from the molecular-recognition benchmark alone
