Literature · OpenMed
OpenMed PharmaDetect
A transformer model for identifying and extracting chemical and drug entities from biomedical and clinical text.
What OpenMed PharmaDetect is
OpenMed PharmaDetect is a token-classification model developed by OpenMed for named entity recognition of chemical compounds, drugs, and therapeutic substances. Built on the SuperClinical backbone using a DeBERTa architecture with 434 million parameters, it is designed to extract pharmaceutical entities from scientific literature and healthcare documents with high recall.
The model addresses the need to automatically detect chemical substances within unstructured medical text. It supports downstream biomedical workflows such as drug interaction detection, adverse event monitoring, medication extraction from patient records, and biomedical knowledge graph construction.
How it works
The model uses the DeBERTa-v3-large architecture for token classification. It takes tokenised biomedical text as input and predicts BIO tags (Beginning, Inside, and Outside) for chemical entities using the B-CHEM and I-CHEM labels.
Training was carried out on the BC5CDR-Chem corpus from the BioCreative V Chemical-Disease Relation extraction task, which contains 1,500 PubMed abstracts and 4,409 annotated chemical entities. The model was trained using the AdamW optimizer with learning rate scheduling and evaluated using cross-validation on a held-out test set.
Key facts
- Base architecture
- deberta-v3-large
- Parameter count
- 434M
- Task
- Token classification (named entity recognition)
- Entity labels
- B-CHEM, I-CHEM
- Training dataset
- BC5CDR-Chem (1,500 PubMed abstracts, 4,409 chemical annotations)
- Optimization
- AdamW with learning rate scheduling
- BC5CDR-Chem F1 score
- 0.9614
- BC5CDR-Chem recall
- 0.9710
- BC5CDR-Chem precision
- 0.9520
- Recommended GPU memory
- 16 GB
What you can use it for
- Adverse event monitoring
- Identifies drug and chemical mentions within clinical documents and research reports to support pharmacovigilance.
- Drug interaction screening
- Extracts co-occurring pharmacological agents from medical literature to assist in identifying potential drug-drug interactions.
- Medication extraction from records
- Parses clinical notes and patient records to extract prescribed medications and therapeutic compounds.
- Biomedical knowledge graph construction
- Pulls chemical entity mentions from PubMed abstracts to populate structured nodes in knowledge bases.
- Literature mining for drug discovery
- Screens published biomedical articles to catalogue chemical compounds for drug repurposing and development studies.
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 PharmaDetect
Strengths and limits
Good at
- Attains an F1 score of 0.9614 and a recall of 0.9710 on the BC5CDR-Chem evaluation set.
- Specifically tuned for chemical and pharmaceutical entity recognition on PubMed abstracts and clinical narratives.
- Compatible with multiple token-level aggregation strategies including simple, first, average, and max.
Watch out for
- Restricted to chemical entities (B-CHEM and I-CHEM) and does not identify diseases, genes, or other biomedical categories.
- Documented training and evaluation are focused specifically on the BC5CDR-Chem corpus.
- Requires 16 GB of GPU memory for standard deployment.
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 PharmaDetect 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 PharmaDetect 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 entities does OpenMed PharmaDetect recognise?
The model recognises chemical compounds, pharmaceutical substances, and therapeutic drugs, labelling them with B-CHEM and I-CHEM tags.
What architecture and backbone does this model use?
It uses the DeBERTa-v3-large architecture configured as the SuperClinical 434M backbone.
What dataset was the model trained on?
It was trained on the BC5CDR-Chem dataset, consisting of 1,500 PubMed abstracts annotated with 4,409 chemical entities from the BioCreative V Chemical-Disease Relation task.
How are subword token predictions grouped into full entity spans?
Predictions can be aggregated using Hugging Face aggregation strategies such as simple tag grouping, or word-level strategies based on the first token, average score, or maximum score.
How to cite it
Maziyar Panahi, 'OpenMed NER: Open-Source, Domain-Adapted State-of-the-Art Transformers for Biomedical NER Across 12 Public Datasets', 2025, arXiv:2508.01630.
Written from OpenMed PharmaDetect'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.