We present BERN2 (Advanced Biomedical Entity Recognition and Normalization), a tool that improves the previous neural network-based NER tool by employing a multi-task NER model and neural network-based NEN models to achieve much faster and more accurate inference. This repository provides a way to host your own BERN2 server. See our paper for more details.
***** Try BERN2 at http://bern2.korea.ac.kr *****
- [Mar 18, 2022] On the web service, we set the API limit of 100 requests per 100 seconds per user. For bulk requests, we highly recommend you to use the local installation.
- [Mar 17, 2022] BERN2 v1.1 has been released. Please see the release page for more information on what's new in this version.
- [Feb 15, 2022] Bioregistry is used to standardize prefixes for normalized entity identifiers.
old | new |
---|---|
MESH:D009369 | mesh:D009369 |
OMIM:608627 | mim:608627 |
CL_0000021 | CL:0000021 |
CVCL_J260 | cellosaurus:CVCL_J260 |
NCBI:txid10095 | NCBITaxon:10095 |
EntrezGene:10533 | NCBIGene:10533 |
- [Jan 07, 2022] BERN2 v1.0 has been released.
You first need to install BERN2 and its dependencies.
# Install torch with conda (please check your CUDA version)
conda create -n bern2 python=3.7
conda activate bern2
conda install pytorch==1.9.0 cudatoolkit=10.2 -c pytorch
conda install faiss-gpu libfaiss-avx2 -c conda-forge
# Check if cuda is available
python -c "import torch;print(torch.cuda.is_available())"
# Install BERN2
git clone [email protected]:dmis-lab/BERN2.git
cd BERN2
pip install -r requirements.txt
(Optional) If you want to use mongodb as a caching database, you need to install and run it.
# https://docs.mongodb.com/manual/tutorial/install-mongodb-on-ubuntu/#install-mongodb-community-edition-using-deb-packages
sudo systemctl start mongod
sudo systemctl status mongod
Then, you need to download resources (e.g., external modules or dictionaries) for running BERN2. Note that you will need 70GB of free disk space. If the link below does not work, please download the resources here.
wget http://nlp.dmis.korea.edu/projects/bern2-sung-et-al-2022/resources_v1.1.tar.gz
tar -zxvf resources_v1.1.tar.gz
rm -rf resources_v1.1.tar.gz
# install CRF
cd resources/GNormPlusJava/CRF
./configure --prefix="$HOME"
make
make install
cd ../../..
The following command runs BERN2.
export CUDA_VISIBLE_DEVICES=0
cd scripts
bash run_bern2.sh
(Optional) To restart BERN2, you need to run the following commands.
export CUDA_VISIBLE_DEVICES=0
cd scripts
bash stop_bern2.sh
bash run_bern2.sh
After successfully running BERN2 in your local environment, you can access it via RESTful API. If you want to use BERN2 without installing it locally, please see here for instructions on how to use the web service.
import requests
def query_plain(text, url="http://localhost:8888/plain"):
return requests.post(url, json={'text': text}).json()
if __name__ == '__main__':
text = "Autophagy maintains tumour growth through circulating arginine."
print(query_plain(text))
import requests
def query_pmid(pmids, url="http://localhost:8888/pubmed"):
return requests.get(url + "/" + ",".join(pmids)).json()
if __name__ == '__main__':
pmids = ["30429607", "29446767"]
print(query_pmid(pmids))
- annotation_v1.0.tar.gz - NER and normalization for 31.8+ millions of PubMed articles (pubmed21n0001 ~ pubmed21n1062 (2021.01.12)) generated by BERN2 v1.0 (Compressed, 20 GB)
The data provided by BERN2 is post-processed and may differ from the most current/accurate data available from U.S. National Library of Medicine (NLM).
@article{sung2022bern2,
title={BERN2: an advanced neural biomedical namedentity recognition and normalization tool},
author={Sung, Mujeen and Jeong, Minbyul and Choi, Yonghwa and Kim, Donghyeon and Lee, Jinhyuk and Kang, Jaewoo},
year={2022},
eprint={2201.02080},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
For help or issues using BERN2, please submit a GitHub issue. Please contact Mujeen Sung (mujeensung (at) korea.ac.kr
), or Minbyul Jeong (minbyuljeong (at) korea.ac.kr
) for communication related to BERN2.