DeepDynaForecast: Phylogenetic-informed graph deep learning for epidemic transmission dynamic prediction

During an outbreak or sustained epidemic, accurate prediction of patterns in transmission risk can reliably inform public health strategies. Projections indicating growth or decline of transmission for specific risk groups can significantly enhance the optimization of interventions, especially when resources are limited. To address this, we present DeepDynaForecast, a cutting-edge deep learning algorithm designed for forecasting pathogen transmission dynamics. Uniquely, DeepDynaForecast was trained on in-depth simulation data, classifying samples according to their dynamics (growth, static, or decline) with accuracy of 91.6%. We evaluated the model’s performance and application using simulated outbreak data and empirical, large-scale data from the HIV epidemic in Florida between 2012 and 2020. We conclude DeepDynaForecast represents a significant advancement in genomics-mediated pathogen transmission characterization and has the potential to catalyze new research directions within virology, molecular biology, and public health.

Paper

This repository provides the official implementation of the model in the following paper:

DeepDynaForecast: Phylogenetic-informed graph deep learning for epidemic transmission dynamic prediction

Chaoyue Sun¹, Ruogu Fang^1,2,3, Marco Salemi^4,5, Mattia Prosperi^5,6* and Brittany Rife Magalis^4,5*

¹ Department of Electrical and Computer Engineering, Herbert Wertheim College of Engineering, University of Florida, Gainesville, Florida, United States of America
² J. Crayton Pruitt Family Department of Biomedical Engineering, Herbert Wertheim College of Engineering, University of Florida, Gainesville, Florida, United States of America
³ Center for Cognitive Aging and Memory, McKnight Brain Institute, University of Florida, Gainesville, Florida, United States of America
⁴ Department of Pathology, Immunology, and Laboratory Medicine, University of Florida, Gainesville, Florida, United States of America
⁵ Emerging Pathogens Institute, University of Florida, Gainesville, Florida, United States of America
⁶ Department of Epidemiology, University of Florida, Gainesville, Florida, United States of America

Dependencies

Please check the dependencies.txt.

Datasets

• Agent-based simulation model nosoi to generate simulated trees and input datatables in R. Please see sim for the simulation codes.
• For Florida HIV epidemic data, please contact FDOH for availabiliy.

Usage examples

• Preparing dataset
  Both original and preprocessed datasets are avaliable. Please uzip them in the root folder. Jupyter notebook files in aly provide a detailed preprocessing step by step for edge feature. The provided preprocessed dataset is divided into train/validation/test sets.
  Before running the code, put the preprocessed dataset on a desired directory. By default, the data root is set as data/preprocessed_data/split_rs123.
  See: dl/config.py

data.add_argument("--ds_name", type=str, default="preprocessed_data", help="The name of dataset")
data.add_argument("--ds_dir", type=str, default="../data/", help="The base folder for data")
data.add_argument("--ds_split", type=str, default="split_rs123")

• Run container
  The container is build by Docker and is avaliable at here. To run a shell within the container:

sudo docker run -it —gpus all -v ddf    

• Train models
  Example commands

# train a DeepDynaForecast model
python main.py --model ddf --model_num 0
# train a DeepDynaForecast model with specific setting
python main.py --model ddf --model_num 0 --batch_size 32 --init_lr 0.001 --min_lr 1e-6 --lr_decay_rate 0.1

Results

To use the well-trained neural network models, please download files and unzip it to dl.
Please change the settings in test/main_test.py to run the test phase with different models.

python main_test.py

Besides, scripts for generating figures in the main pages and supplementaries are avaliable at the test folder.

Citation

If you use this code, please cite our papers:

@article{sun2024deepdynaforecast,
  title={DeepDynaForecast: Phylogenetic-informed graph deep learning for epidemic transmission dynamic prediction},
  author={Sun, Chaoyue and Fang, Ruogu and Salemi, Marco and Prosperi, Mattia and Rife Magalis, Brittany},
  journal={PLOS Computational Biology},
  volume={20},
  number={4},
  pages={e1011351},
  year={2024},
  publisher={Public Library of Science San Francisco, CA USA}
}

Acknowledgement

The authors acknowledge University of Florida Research Computing for providing computational resources and support that have contributed to the research results reported in this publication.