Literature · OpenMed
OpenMed ChemicalDetect
A 149M-parameter transformer for detecting chemical compounds and drug names in biomedical and clinical text.
What OpenMed ChemicalDetect is
OpenMed ChemicalDetect is a fine-tuned token-classification model built to locate and label chemical compounds, drugs, and other chemical substances in biomedical and clinical writing. Developed by OpenMed and based on the ModernMed 149M backbone, it identifies chemical spans to support literature mining, pharmacovigilance, and automated medical record processing.
The model tags entities using the standard BIO schema, identifying the start and continuation of chemical spans. It addresses the challenge of accurately extracting complex chemical nomenclature from research abstracts and healthcare documents, which is a foundational requirement for building biomedical knowledge graphs and monitoring adverse drug events.
How it works
The model uses the ModernBERT-base transformer architecture configured for token classification. It was trained using the Hugging Face Transformers framework with the AdamW optimizer and a learning rate schedule, with validation conducted via cross-validation on a held-out test split.
Fine-tuning was conducted on the BC4CHEMD corpus from the BioCreative IV challenge. This dataset comprises PubMed abstracts manually annotated with chemical substances according to Chemical Entities of Biological Interest (ChEBI) guidelines, enabling the model to learn chemical terminology from published scientific literature.
Key facts
- Architecture
- ModernBERT-base
- Parameter count
- 149M
- Task
- Token classification (Named Entity Recognition)
- Training dataset
- BC4CHEMD
- Input modality
- Biomedical text
- Output format
- BIO-tagged entity predictions (B-CHEM, I-CHEM)
- F1 score
- 0.92
- Precision
- 0.91
- Recall
- 0.93
- Accuracy
- 0.98
What you can use it for
- Literature Mining for Drug Discovery
- Extracts chemical compounds, experimental substances, and approved drug names from PubMed abstracts and research articles.
- Medication Extraction from Clinical Records
- Parses electronic health documents and patient notes to detect administered drugs and active ingredients.
- Drug Interaction and Adverse Event Monitoring
- Surveys scientific literature and healthcare texts to identify co-occurring chemical entities for pharmacovigilance workflows.
- Biomedical Knowledge Graph Construction
- Identifies chemical mentions across text corpora to serve as structured chemical nodes in relational biomedical graphs.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
OpenMed ChemicalDetect
Strengths and limits
Good at
- Attains high precision (0.91) and recall (0.93) on chemical entity mentions in benchmark evaluation.
- Fine-tuned on the standardized BC4CHEMD dataset, aligning annotations with established ChEBI guidelines.
- Utilises a lightweight 149M-parameter ModernMed backbone, suitable for batched local inference on standard hardware.
Watch out for
- Trained exclusively for chemical entity recognition and cannot recognise other biomedical entities such as genes, diseases, or cell lines.
- Only emits B-CHEM and I-CHEM tags, without grounding identified spans to external database identifiers like ChEBI IDs or PubChem CIDs.
- Achieves a lower F1 score (0.92) on the BC4CHEMD benchmark compared to larger variants in the OpenMed chemical suite, which reach up to 0.954.
Inputs and outputs
You provide
Text
Entities are labelled token by token.
You get back
Labels with confidence scores, highest first.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use OpenMed ChemicalDetect commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run OpenMed ChemicalDetect without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
What entity tags does OpenMed ChemicalDetect output?
The model outputs B-CHEM and I-CHEM tags, representing the beginning and continuation of chemical entity mentions respectively.
On which dataset was this model trained?
It was fine-tuned on the BioCreative IV Chemical Entity Mention (BC4CHEMD) corpus, which contains PubMed abstracts annotated according to ChEBI guidelines.
Can this model normalize entities to standard database identifiers like ChEBI or PubChem?
No. The model performs token classification to extract chemical entity spans from text, but does not perform entity linking or concept normalization.
What architecture does the 149M model use?
It is built on top of the ModernBERT-base architecture (ModernMed 149M backbone) fine-tuned for token classification.
How to cite it
Maziyar Panahi. OpenMed NER: Open-Source, Domain-Adapted State-of-the-Art Transformers for Biomedical NER Across 12 Public Datasets. arXiv:2508.01630 [cs.CL], 2025.
Written from OpenMed ChemicalDetect's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.