Home » Notes » Datasets » Document Processing Datasets

MolParser Datasets

MolParser-7M and WildMol Datasets for Robust Chemical Structure Recognition
2025-10-03 · Hunter Heidenreich
Dataset Details
AuthorsXi Fang, Jiankun Wang, Xiaochen Cai, Shangqian Chen, Shuwen Yang, Haoyi Tao, Nan Wang, Lin Yao, Linfeng Zhang, Guolin Ke
Paper TitleMolParser: End-to-end Visual Recognition of Molecule Structures in the Wild
InstitutionDP Technology
Published InarXiv
CategoryDocument Processing
FormatMolecule Images (PNG) Extended SMILES (E-SMILES) strings
SizeTest molecules: 20,000
Training pairs: 7,740,871
DateOctober 2025
Year2025
Links📊 Dataset • 📄 Paper
A colored molecule with annotations, representing the diverse drawing styles found in scientific papers that OCSR models must handle.
The MolParser system is designed to perform end-to-end recognition of molecular structures found in real-world documents like patents and scientific literature.

Key Contribution

Introduces MolParser-7M, the largest Optical Chemical Structure Recognition (OCSR) dataset, uniquely combining diverse synthetic data with a large volume of manually-annotated, ‘in-the-wild’ images from real scientific documents to improve model robustness. Also introduces WildMol, a new challenging benchmark for evaluating OCSR performance on real-world data, including Markush structures.

Dataset Information

Format

Molecule Images (PNG) Extended SMILES (E-SMILES) strings

Size

TypeCount
Test Molecules20,000
Training Pairs7,740,871

Dataset Examples

An example of a complex Markush structure that can be represented by the E-SMILES format but not by standard SMILES or FG-SMILES.
An example of a complex Markush structure that can be represented by the E-SMILES format but not by standard SMILES or FG-SMILES.
A sample from the WildMol benchmark, showing a low-quality, noisy molecular image cropped from real-world literature that challenges OCSR systems.
A sample from the WildMol benchmark, showing a low-quality, noisy molecular image cropped from real-world literature that challenges OCSR systems.
A colored molecule with annotations, representing the diverse drawing styles found in scientific papers that OCSR models must handle.
A colored molecule with annotations, representing the diverse drawing styles found in scientific papers that OCSR models must handle.

Dataset Subsets

SubsetCountDescription
MolParser-7M (Training Set)7,740,871A large-scale dataset for training OCSR models, split into pre-training and fine-tuning stages.
WildMol (Test Set)20,000A benchmark of 20,000 human-annotated samples cropped from real PDF files to evaluate OCSR models in ‘in-the-wild’ scenarios.

Results

Optical Chemical Structure Recognition ( Accuracy %)

ModelWildMol-10K
OSRA 2.126.3
MolVec 0.9.726.4
Imago 2.06.9
Img2Mol24.4
MolGrapher45.5
DECIMER 2.756.0
MolScribe🥈 66.4
MolParser-Base🥇 76.9

Strengths

  • Largest open-source OCSR dataset with over 7.7 million pairs.
  • The only large-scale OCSR training set that includes a significant amount (400k) of ‘in-the-wild’ data cropped from real patents and literature.
  • High diversity of molecular structures from numerous sources (PubChem, ChEMBL, polymers, etc.).
  • Introduces the WildMol benchmark for evaluating performance on challenging, real-world data, including Markush structures.
  • The ‘in-the-wild’ fine-tuning data (MolParser-SFT-400k) was curated via an efficient active learning data engine with human-in-the-loop validation.

Limitations

  • The E-SMILES format cannot represent certain complex cases, such as coordination bonds, dashed abstract rings, and Markush structures depicted with special patterns.
  • The model and data do not yet fully exploit molecular chirality, which is critical for chemical properties.
  • Performance could be further improved by scaling up the amount of real annotated training data.

Technical Notes

Synthetic Data Generation

  • To ensure diversity, molecular structures were collected from databases like ChEMBL, PubChem, and Kaggle BMS.
  • A significant number of Markush, polymer, and fused-ring structures were also randomly generated.
  • Images were rendered using RDKit and epam.indigo with randomized parameters (e.g., bond width, font size, rotation) to increase visual diversity.

In-the-Wild Data Engine (MolParser-SFT-400k)

  • A YOLOv11 object detection model (MolDet) located and cropped over 20 million molecule images from 1.22 million real PDFs (patents and papers).
  • After de-duplication via p-hash similarity, 4 million unique images remained.
  • An active learning algorithm was used to select the most informative samples for annotation, targeting images where an ensemble of models showed moderate confidence (0.6-0.9 Tanimoto similarity), indicating they were challenging but learnable.
  • This active learning approach with model pre-annotations reduced manual annotation time per molecule to 30 seconds, a 90% savings compared to annotating from scratch.