PharmaHKG-Text

The implementation of Representing a Heterogeneous Pharmaceutical Knowledge-Graph with Textual Information.

Requirements

pip3 install -r requirements.txt

Usage

Constructing dataset

Data files of DrugBank, UniProt, MeSH and SMPDB can be freely donwloaded from the official website. (You need to create an account to download DrugBank file.)

export DRUGBANK_DATA=(DrugBank XML file path)
export UNIPROT_DATA=(UniProt Swiss-Prot XML file path)
export MESH_DATA=(MeSH description XML file path)
export SMPDB_DATA=(SMPDB TSV file path)

sh construct_kg.sh

Preparing BERT embeddings

sh prepare_bert_embs.sh

Link prediction with textual embeddings

cd dgl-ke/python/dglke

python3 train.py \
  --model_name SimplE \
  --dataset PharmaHKG \
  --data_path ../../../dataset/noaug \
  --data_file train.tsv valid.tsv test.tsv \
  --format 'raw_udd_hrt' \
  --batch_size 4096 \
  --log_interval 1000 \
  --neg_sample_size 1 \
  --hidden_dim 768 \
  --lr 0.25 \
  --neg_sample_size_eval -1 \
  --negative_sampling_for_hetero \
  --loss_genre Logistic \
  --gpu 0 \
  --mix_cpu_gpu \
  --num_epochs 100 \
  --valid \
  --eval_interval_epoch 100 \
  --save_path OUTPUT_DIR \
  --entity_emb_file noaug_name \
  --async_update

When you use the (B) Alignemnt method, please add --do_alignment. When using the (C) Augmentation method, please set --data_path to dataset/aug and set the ---entity_emb_file to aug.

Citation

@ARTICLE{10.3389/frma.2021.670206,
AUTHOR={Asada, Masaki and Gunasekaran, Nallappan and Miwa, Makoto and Sasaki, Yutaka},   
TITLE={Representing a Heterogeneous Pharmaceutical Knowledge-Graph with Textual Information},      
JOURNAL={Frontiers in Research Metrics and Analytics},      
VOLUME={6},      
PAGES={39},     
YEAR={2021},      
URL={https://www.frontiersin.org/article/10.3389/frma.2021.670206},       
DOI={10.3389/frma.2021.670206},      
ISSN={2504-0537},   
}

Acknowledgement

This work was supported by JSPS KAKENHI Grant Number 20k11962.