BIOS_EntityClassification

Contents

• Overview
• Repo Contents
• System Requirements
• Installation Guide
• Demo
• License
• Issues
• Citation

Overview

A Semantic Type Annotator(STA) to predict the semantic type of a medical term using the term itself as well as its surrounding text as input. Training samples were multi-word terms from the UMLS, as they generally do not have ambiguous daily meaning. The semantic types of these terms were used as the label, and in case a term has multiple semantic types in the UMLS, a random one is used, leveraging the fact that a large sample size can overcome moderate noise in the data. The classification model was trained on PubMedBERT, a BERT model pretrained on PubMed abstracts.

Repo Contents

• pretrain: scripts for downloading pretrained models
• example: small dataset to demo the code
• train: train codes
• predict: predict codes
• utils: utils codes

System Requirements

Hardware Requirements

For optimal performance, we recommend a computer with following specs:

RAM: 16+ GB
CPU: 4+ cores, 3.3+ GHz/core
GPU Memory: 40 GB

Software requirements

The package is tested on Linux 20.04 operating system.

Install requirements

!pip install -r requirements.txt

Installation Guide

1. Download cleanterms5.txt (password: d9ol) and put it under ./example/cleanterms/.
2. Download pretrained models: cd pretrain && sh download_pubmedbert.sh
   which will take about a few minutes to complete the download.

Demo

Train

make sure cleanterms5.txt has been downloaded, and then:

1. cd train
2. bash train.sh

after a few minutes, you can find your fine-tuned model and eval result under ./output_xxx by default.

Predict

configure your fine-tuned model path in the predict.sh, and then:

1. cd predict
2. bash predict.sh

after a few minutes, you can find the results under ./output_xxx

Instructions for use

1. generate your cleanterms.txt from UMLS by your rules. and python utils/label_util.py to generate the entity_type.json and entity_group.json.

2. prepare your training texts and the FMM results.

3. use your datasets to train and predict!

Citation

@misc{https://doi.org/10.48550/arxiv.2203.09975,
  doi = {10.48550/ARXIV.2203.09975},
  url = {https://arxiv.org/abs/2203.09975},
  author = {Yu, Sheng and Yuan, Zheng and Xia, Jun and Luo, Shengxuan and Ying, Huaiyuan and Zeng, Sihang and Ren, Jingyi and Yuan, Hongyi and Zhao, Zhengyun and Lin, Yucong and Lu, Keming and Wang, Jing and Xie, Yutao and Shum, Heung-Yeung},
  keywords = {Computation and Language (cs.CL), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {BIOS: An Algorithmically Generated Biomedical Knowledge Graph},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International}
}