Literature · OpenMed
OpenMed ProteinDetect
A 141M-parameter DeBERTa model that identifies proteins, complexes, and variants in biomedical literature and clinical text.
What OpenMed ProteinDetect is
OpenMed ProteinDetect (OpenMed-NER-ProteinDetect-SuperClinical-141M) is a token-classification model developed by OpenMed for biomedical named entity recognition (NER). It is designed to identify and extract mentions of proteins, protein complexes, protein families or groups, and protein variants from scientific articles and clinical documents.
Manual extraction of molecular entities from expanding scientific literature and patient documentation presents a major bottleneck for bioinformatics and systems biology. By automating the detection of specific molecular biology entities, this model aids tasks such as protein–protein interaction extraction, molecular pathway mapping, and biomedical knowledge graph construction.
How it works
The model is based on the DeBERTa-v3-small architecture (tagged as deberta-v2 in repository metadata) and contains 141 million parameters. It accepts tokenised biomedical text and outputs token-level predictions using the standard Begin-Inside-Outside (BIO) tagging scheme.
The model was fine-tuned on the Florida State University (FSU) corpus, a specialised biomedical NER dataset tailored for protein interactions and molecular biology entities. Training was executed using Hugging Face Transformers and the AdamW optimiser with learning rate scheduling, evaluated via cross-validation on a held-out test set.
Key facts
- Architecture
- DeBERTa-v3-small
- Parameters
- 141 million (141M)
- Task
- Token classification (Named Entity Recognition)
- Training data
- FSU (Florida State University) corpus
- Input modality
- English text
- Output format
- BIO-tagged entity tokens with confidence scores
- Overall F1 score
- 0.92
- Overall precision
- 0.91
- Overall recall
- 0.93
- Overall accuracy
- 0.97
What you can use it for
- Protein-protein interaction mining
- Extract individual protein names, complexes, and variants from published studies to automate the identification of physical or functional interactions.
- Biomedical knowledge graph construction
- Identify molecular biology entities across large text corpora to populate nodes and relations for biological pathways and signal transduction networks.
- Clinical literature and records mining
- Detect protein entities, family groups, and variants mentioned within healthcare documents and medical texts for systems biology research.
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 ProteinDetect
Strengths and limits
Good at
- Demonstrates strong extraction performance on its target domain, achieving an F1 score of 0.92, recall of 0.93, and accuracy of 0.97 on the FSU benchmark.
- Distinguishes between fine-grained molecular biology classes, including individual proteins, protein complexes, protein groups or families, and protein variants.
- Provides a compact 141M-parameter footprint suitable for high-throughput batch processing on consumer or single GPU hardware.
Watch out for
- Limited to English-language biomedical and clinical texts.
- Trained specifically on the FSU corpus entity taxonomy and may not generalise to broader biomedical concepts such as diseases, chemicals, or anatomical sites.
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 ProteinDetect 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 ProteinDetect 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 extract?
The model identifies five main categories formatted in BIO tags: individual proteins (protein), protein complexes (protein_complex), protein enumerations (protein_enum), protein families or groups (protein_familiy_or_group), and protein variants (protein_variant).
What dataset was used to train the model?
The model was trained on the Florida State University (FSU) corpus, a biomedical dataset designed for extracting molecular biology entities and protein interactions in systems biology and biochemistry.
What aggregation strategies are recommended for inference?
When using the Hugging Face pipeline, users can apply aggregation strategies such as 'none' for raw tokens, 'simple' to merge adjacent tokens of the same entity class, or word-level strategies like '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. arXiv:2508.01630 [cs.CL], 2025.
Written from OpenMed ProteinDetect'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.