Medical history predicts phenome-wide disease onset and enables the rapid response to emerging health threats
This repository contains the implementation of the deep learning model used in "Medical history predicts phenome-wide disease onset and enables the rapid response to emerging health threats".
The subfolder ehrgraphs contains the repository used for the models used in
the publication. Included is a environment.yml file for setting up the conda
environment used for all experiments, a setup.py allowing for installation of
the package via pip install ehrgraphs with required dependencies.
To reproduce the results you will need to have access to UK Biobank and use the preprocessing code in this repository: https://github.com/JakobSteinfeldt/22_medical_records
We also provide a standalone implementation with synthetic data in that can be used to train a model similar to the one used in the publication, but without much of the complexity of the original implementation.
We recommend adapting this standalone implementation if you are trying to build a similar model for your data.
The standalone implementation can be found in medical_history_model_standalone.ipynb.
The models were trained on HPC nodes with 64 cpu cores, 512 GB RAM, and one NVIDIA A100 80G GPU.
tbd