Drug Repuprosing using Consilience of Knowledge Graph Completion Methods

This repository is a fork of Sun et. al implementation of four knowledge graph embedding models. Here we apply the aforementioned algorithms to a biomedical knowledge graph called MIND (MechRepoNet with DrugCentral indications). We report the results of our analysis in this preprint.

Modifications to the original repository

• Added code to output raw embeddings in order to extract predictions. This can be seen with the --do_predict flag in codes/run.py.
• Added Notebooks folder that encapsulates analysis done on the MIND dataset.
• Added methods, Notebooks/score_utils.py, to process and translate raw embeddings into human readable entities and relations.

Usage instructions

1. Please see the original PyTorch implementation instructions
2. Download the MIND dataset to ./data
3. Install requirements into python virtual environment

   # run in shell
   mamba create -f environment.yml
   mamba activate kge
   
   

4. Train/Test

   # run in shell
   bash run.sh train <model_name> <dataset_name> <gpu_num> <folder_out_name> <batch size> <neg_sample_size> <dimensions> <gamma> <alpha> <learningrate> <test_batch_size> <double_entities_emb> <double_relation_emb> <regularization>
   
   # or in python
   # for more parameters please see codes/run.py Lines 23 - 72
   python run.py --{do_train, do_valid, do_test, do_predict} --data_path <where/data/is> --model {TransE, DistMult, ComplEx, RotatE}
   

Citation

If you use the codes, please cite the original paper by Sun et al:

@inproceedings{
 sun2018rotate,
 title={RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space},
 author={Zhiqing Sun and Zhi-Hong Deng and Jian-Yun Nie and Jian Tang},
 booktitle={International Conference on Learning Representations},
 year={2019},
 url={https://openreview.net/forum?id=HkgEQnRqYQ},
}