Literature · OpenMed
OpenMed DNADetect
A 125M-parameter RoBERTa model for extracting DNA, RNA, protein, cell line, and cell type entities from biomedical text.
What OpenMed DNADetect is
OpenMed DNADetect (OpenMed-NER-DNADetect-SuperMedical-125M) is a specialised token-classification model developed by OpenMed for biomedical named entity recognition (NER). It identifies references to molecular biology entities, focusing on gene-level DNA sequences, RNA, proteins, cell lines, and cell types in scientific literature and clinical texts.
Manual extraction of molecular entities from the expanding biomedical literature is labour-intensive. By automating the identification of structural and functional biological mentions in text, the model supports applications such as literature mining for drug discovery, biomedical knowledge graph construction, and clinical document analysis.
How it works
The model uses a RoBERTa-base transformer backbone with approximately 125 million parameters configured for token classification. It processes tokenised biomedical text and outputs token-level predictions using a beginning-inside-outside (BIO) tagging scheme across five entity classes: DNA, RNA, protein, cell line, and cell type.
It was fine-tuned on the JNLPBA corpus, a benchmark derived from GENIA MEDLINE abstracts annotated for the 2004 bio-entity recognition task. Fine-tuning was carried out within the Hugging Face Transformers framework using the AdamW optimiser with learning rate scheduling, evaluated through cross-validation on a held-out test set.
Key facts
- Base architecture
- roberta-base
- Parameter count
- 125M
- Task
- Token classification (named entity recognition)
- Fine-tuning dataset
- JNLPBA
- Entity categories
- DNA, RNA, protein, cell line, cell type
- Evaluation F1 score
- 0.8044
- Evaluation precision
- 0.7589
- Evaluation recall
- 0.8557
- Evaluation accuracy
- 0.9284
What you can use it for
- Biomedical literature mining
- Extract mentions of genes, DNA sequences, RNAs, and proteins across MEDLINE abstracts and research articles to assist in functional genomics and drug discovery research.
- Knowledge graph construction
- Detect biological entities such as proteins, cell types, and cell lines from unstructured scientific text to populate structured biomedical databases and relation networks.
- Clinical documentation analysis
- Identify biological entities and molecular factors referenced in clinical texts, patient records, or adverse event monitoring pipelines.
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 DNADetect
Strengths and limits
Good at
- Demonstrates higher recall (0.8557) relative to precision (0.7589) on the JNLPBA benchmark, reducing the rate of missed entity mentions.
- Detects five distinct biological entity classes using standard BIO tagging: DNA, RNA, protein, cell line, and cell type.
- Compatible with standard Hugging Face token-classification pipelines, enabling simple integration with entity aggregation strategies.
Watch out for
- Lower overall F1 score (0.8044) on the JNLPBA benchmark compared to larger variants in the OpenMed DNADetect family, such as SuperClinical-434M and SuperMedical-355M.
- Limited to five specific molecular entity classes and does not natively tag other biomedical concepts like diseases or chemical compounds.
- Requires downstream token aggregation strategies (such as simple, average, or max) to convert subword token labels into full entity spans.
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 DNADetect 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 DNADetect 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 types can OpenMed DNADetect identify?
It detects five entity classes using BIO tags: DNA (B-DNA, I-DNA), RNA (B-RNA, I-RNA), protein (B-protein, I-protein), cell line (B-cell_line, I-cell_line), and cell type (B-cell_type, I-cell_type).
What data was this model trained on?
The model was fine-tuned on the JNLPBA corpus, a dataset derived from MEDLINE abstracts annotated for the 2004 bio-entity recognition task.
How does this model handle token aggregation into entities?
The model outputs raw BIO-tagged token predictions. In a Hugging Face pipeline, grouping into complete entity spans can be controlled using aggregation strategies such as simple, first, average, or max.
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 [cs.CL].
Written from OpenMed DNADetect'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.