KindMed: Knowledge-Induced Medicine Prescribing Network for Medication Recommendation

This repository provides the implementation of KindMed as the medical knowledge graph-driven medicine recommender framework.

Dependencies

Our implementation mainly utilized PyTorch 1.11.0 and PyTorch Geometric 2.2.0. Additional tensorflow/tensorboard libraries was used for training logging.

torch==1.11.0
torch-scatter
torch-sparse
torch-cluster
torch-spline-conv
torch-geometric==2.2.0
tensorboard==2.11.2
tensorflow-gpu==2.11.0

Training

For training KindMed, run the following code:

python main_kindmed.py --gpu_id=0 --phase='training'

Evaluation

For testing the KindMed, provide the optimized model on 'path_to_model' variable, and run the following code:

python main_kindmed.py --gpu_id=0 --phase='testing'

Code Details

• main_kindmed.py : the main code to execute the proposed model for training / testing based on the given argument(s).
• models.py : the code that defines the model of KindMed and the fusion module
• losses.py : the code that defines the DDI loss to train the model
• helpers.py : the code that defines a set of helper functions for running the model

Datasets

We utilized MIMIC-III and MIMIC-IV datasets as our EHR main cohorts.

• https://physionet.org/content/mimiciii/
• https://physionet.org/content/mimiciv/

Citation

If you find this work useful for your research, please cite our preprint paper.

@misc{mulyadi2022xadlime,
  doi={10.48550/arXiv.2310.14552},
  url={https://arxiv.org/abs/2310.14552},
  author={Mulyadi, Ahmad Wisnu and Suk, Heung-Il},
  title={Knowledge-Induced Medicine Prescribing Network for Medication Recommendation},
  publisher={arXiv},
  year={2022},
  copyright={arXiv.org perpetual, non-exclusive license}
}

Acknowledgements

This work was supported by National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) No. 2022R1A2C2006865 (Development of deep learning techniques for data-driven medical knowledge graph generation and interpretable multi-modal electronic health records analysis). This work was further supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) No. 2019-0-00079 (Artificial Intelligence Graduate School Program (Korea University)) and No. 2022-0-00959 ((Part 2) Few-Shot Learning of Causal Inference in Vision and Language for Decision Making).