HowML/biobert-ner-bionlp13pc-fine-tuned

A token-classification model fine-tuned on the BioNLP 2013 Pathway Curation (BioNLP13PC) corpus for biomedical named entity recognition. The model tags mentions of chemicals, gene/protein products, complexes, and cellular components in scientific text.

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline

model_name = "HowML/bert-ner-bionlp13pc-fine-tuned"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
print(model.id2label)

"""
{
"0": "O",
"1": "B-CHED",
"2": "B-COMP",
"3": "B-Complex",
"4": "B-PRGE",
"5": "I-CHED",
"6": "I-COMP",
"7": "I-Complex",
"8": "I-PRGE"
}
"""

tokenized_inputs = tokenizer(text, return_tensors="pt")
outputs = model(**tokenized_inputs)
predictions = torch.argmax(outputs.logits, dim=-1)

"""
example output:
[CLS] AKT regulate NF ##kappa ##B [SEP]
predictions = tensor([[0, 4, 0, 4, 8, 8, 0]])
"""

Model Details

Task	Named Entity Recognition (token classification)
Corpus	BioNLP13PC (BioNLP Shared Task 2013 — Pathway Curation)
Base model	dmis-lab/biobert-base-cased-v1.1
Language	English (biomedical)
Labeling scheme	BIO (B-, I-, O)

Entity types

Tag	Entity	Description
CHED	Chemical / Drug	Simple chemical entities
PRGE	Gene / Protein	Genes and gene products (proteins)
COMP	Cellular Component	Subcellular locations and structures
Complex	Complex	Macromolecular complexes

Performance

Evaluated on the BioNLP13PC test set (entity-level, span-based).

Entity	Precision	Recall	F1	Support
CHED	0.83	0.86	0.84	857
COMP	0.88	0.89	0.88	332
Complex	0.75	0.75	0.75	528
PRGE	0.87	0.91	0.89	3586
micro avg	0.85	0.89	0.87	5303
macro avg	0.83	0.85	0.84	5303
weighted avg	0.85	0.89	0.87	5303