Literature · OpenMed
OpenMed GenomeDetect
A 149M-parameter ModernBERT model for extracting gene and protein entity mentions from English biomedical text.
What OpenMed GenomeDetect is
OpenMed GenomeDetect (OpenMed-NER-GenomeDetect-ModernMed-149M) is a token-classification transformer developed by OpenMed and Maziyar Panahi. It targets named entity recognition (NER) in biomedical text, specifically locating and classifying mentions of genes and proteins using standard BIO formatting.
Automated identification of gene and protein terminology is often hindered by complex nomenclature and ambiguous span boundaries. By training on curated literature corpora, the model provides automated extraction of biological entities from research articles, clinical reports, and healthcare documentation.
The extracted mentions support downstream biological and translational workflows, including literature mining for functional genomics, biomedical knowledge graph construction, adverse event monitoring, and candidate identification in drug discovery pipelines.
How it works
The model is built upon the ModernBERT-base architecture (149 million parameters) adapted for token classification. It processes tokenised biomedical English text and outputs token-level predictions corresponding to beginning (`B-GENE/PROTEIN`) and inside (`I-GENE/PROTEIN`) entity tags.
Fine-tuning was conducted on the BioCreative II Gene Mention (BC2GM) corpus, which comprises thousands of sentences annotated for gene and protein names from MEDLINE abstracts. Training was carried out with the Hugging Face Transformers framework using the AdamW optimiser and learning rate scheduling, evaluated by cross-validation on a held-out test set.
Key facts
- Base architecture
- ModernBERT-base
- Parameter count
- 149M
- Primary task
- Token classification (Named Entity Recognition)
- Entity types
- B-GENE/PROTEIN, I-GENE/PROTEIN
- Training dataset
- BioCreative II Gene Mention (BC2GM)
- Evaluation F1 score
- 0.81
- Evaluation precision
- 0.80
- Evaluation recall
- 0.81
- Evaluation accuracy
- 0.95
- Input modality
- English text
What you can use it for
- Gene and protein mention extraction
- Identifies spans referring to genes and proteins across MEDLINE abstracts, scientific publications, and clinical documentation.
- Biomedical knowledge graph assembly
- Extracts gene and protein entities from literature to serve as nodes for biological relationship networks and automated knowledge graphs.
- Drug discovery and target identification
- Mines large collections of scientific literature to identify gene and protein targets relevant to therapeutic mechanisms.
- Clinical text and records mining
- Parses medical documents and patient notes to detect mentions of diagnostic genes or therapeutic protein biomarkers.
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 GenomeDetect
Strengths and limits
Good at
- Demonstrates an overall accuracy of 0.95 and an F1 score of 0.81 on the standard BC2GM evaluation benchmark.
- Leverages the ModernBERT-base architecture for lightweight 149M-parameter inference.
- Outputs standardised BIO tags that can be readily post-processed using standard token aggregation strategies (such as simple, first, average, or max grouping).
Watch out for
- Performance on BC2GM (F1: 0.81) is lower than larger OpenMed variants, such as SuperClinical-434M (F1: 0.9010) or PubMed-335M (F1: 0.8963).
- Restricted to identifying gene and protein entities and cannot distinguish separate subcategories beyond the unified GENE/PROTEIN label.
- Only supports English-language textual inputs.
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 GenomeDetect 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 GenomeDetect 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 categories can this model identify?
The model detects gene and protein mentions, outputting `B-GENE/PROTEIN` for the start of an entity and `I-GENE/PROTEIN` for subsequent continuation tokens.
On which dataset was this model trained?
It was trained and evaluated on the BioCreative II Gene Mention (BC2GM) dataset, which consists of thousands of annotated sentences from MEDLINE abstracts.
How does this model compare to other OpenMed GenomeDetect models?
At 149M parameters, it achieves an F1 score of 0.81, whereas larger variants in the series (ranging from 108M to 568M parameters) reach BC2GM F1 scores between 0.8833 and 0.9010.
What inputs does the model expect?
It accepts raw or tokenised biomedical text strings and outputs entity spans accompanied by confidence scores.
How to cite it
Panahi, M. (2025). OpenMed NER: Open-Source, Domain-Adapted State-of-the-Art Transformers for Biomedical NER Across 12 Public Datasets. arXiv:2508.01630 [cs.CL].
Written from OpenMed GenomeDetect'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.