Literature · OpenMed
OpenMed OncologyDetect
A biomedical transformer for named entity recognition of cancer genetics and anatomical entities in text.
What OpenMed OncologyDetect is
OpenMed OncologyDetect (openmed-openmed-ner-oncologydetect-multimed-568m) is a specialised token-classification model developed by OpenMed and Maziyar Panahi. It targets cancer genetics by identifying mentions of genes, pathological formations, chemicals, anatomical structures, and cancer types within clinical and biomedical literature.
Text mining in oncology requires accurate extraction of biological entities spanning multiple levels of organisation, from molecular factors such as amino acids and gene products to tissues, organs, and disease processes. This model automates the extraction of these entities from scientific papers and healthcare records.
The model assists researchers by facilitating automated literature mining, identifying candidate therapeutic targets, assisting adverse event monitoring, and extracting structured information to construct biomedical knowledge graphs.
How it works
The model is built on the bge-m3 base transformer architecture (associated with XLM-RoBERTa) and fine-tuned for token classification using the Hugging Face Transformers framework. Optimisation is performed with the AdamW optimizer alongside learning rate scheduling.
Training and evaluation are conducted on the BioNLP 2013 CG (Cancer Genetics) corpus, which contains annotations focused on cancer-related genetic entities, gene regulation, and tumor biology. The model outputs token-level predictions using the BIO (Beginning, Inside, Outside) tagging scheme across 16 core entity categories, achieving an overall F1 score of 0.8766 on the dataset.
Key facts
- Base architecture
- bge-m3 (XLM-RoBERTa)
- Model size
- 568M parameters
- Task
- Token classification / Named Entity Recognition (NER)
- Modality
- Text
- Primary training dataset
- BioNLP 2013 CG (Cancer Genetics)
- Language
- English
What you can use it for
- Cancer literature mining
- Extract mentions of oncogenes, tumor suppressor genes, and cancer types from biomedical research articles.
- Biomedical knowledge graph construction
- Identify links between simple chemicals, gene products, cellular components, and pathological formations to populate structured biological databases.
- Clinical text mining
- Detect anatomical systems, tissues, and cancer-related diagnostic terminology from medical reports and healthcare records.
- Therapeutic target and drug interaction discovery
- Surface relationships between chemical compounds, amino acids, and cancer-associated genes for oncology drug development.
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 OncologyDetect
Strengths and limits
Good at
- Achieves an F1 score of 0.8766, a precision of 0.8749, and an accuracy of 0.9351 on the BioNLP 2013 CG benchmark.
- Identifies 16 distinct biological entity classes across cellular, anatomical, genetic, and pathological levels.
- Integrates into token-classification pipelines supporting various entity grouping and aggregation strategies.
Watch out for
- Requires substantial GPU memory (recommended 16 GB) for efficient batch inference.
- Entity identification is restricted to categories defined within the BioNLP 2013 CG annotation scheme.
- Ranked fifth in F1 performance among the OpenMed OncologyDetect suite, scoring lower on the BioNLP 2013 CG benchmark than variants such as SuperMedical-355M and ElectraMed-560M.
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 OncologyDetect 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 OncologyDetect 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 this model identify?
The model detects 16 distinct entity types using BIO tagging: Amino_acid, Anatomical_system, Cancer, Cell, Cellular_component, Developing_anatomical_structure, Gene_or_gene_product, Immaterial_anatomical_entity, Multi-tissue_structure, Organ, Organism, Organism_subdivision, Organism_substance, Pathological_formation, Simple_chemical, and Tissue.
What is the underlying model architecture?
The model uses the 568-million parameter bge-m3 architecture, an XLM-RoBERTa-based transformer, adapted with a token-classification head.
How should multi-token entities be grouped during inference?
Predictions can be grouped using Hugging Face aggregation strategies such as 'simple', 'first', 'average', or 'max' to combine subtoken and multi-token spans.
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 OncologyDetect'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.