A physiologically based digital twin for alcohol consumption – predicting real-life drinking responses and long-term plasma PEth
This repository contains the code for the model developed in the publication titled "A physiologically based digital twin for alcohol consumption – predicting real-life drinking responses and long-term plasma PEth", published in npj Digital Medicine.
The model describe the appearance of alcohol in blood [mg/dL], breakdown to acetaldehyde, and conversion to phosphatidylethanol (PEth).
The model files used to generate the results are available in Matlab/Models, with the primary model being Matlab/Models/Alcoholmodel.txt (available here). These files contain the model equations in a plain text format.
We implemented the model in two different frameworks to make it more accessible. The primary work was done in MATLAB using IQM tools by IntiQuan, but we also provide a Python implementation in a custom toolbox.
The MATLAB implementation is using IQM tools, which is a package that interfaces to the SUNDIALS suite of solvers.
To run the scripts which generate the plots in the paper, run Matlab/alcoholModel.m. Note that you need to have MATLAB installed with a working C-compiler.
To plot the supplementary figures, run MATLAB/plotSupplementaryFigures.m.
For transparency, we also include the scripts used to run the parameter optimization. These can be run by passing extra arguments to alcoholModel: alcoholModel(estimateOnAllData, doOptimize, doPI, doMCMC, doPPL).
The additional scripts and functions used to run the simulations are available in the Matlab/scripts folder.
The Python implementation was implemented using a custom toolbox inspired by the IQM tools. The SUND (SimUlation of Nonlinear Dynamic models) toolbox is still in development, but an initial version is available here. To install all packages needed, the simplest way is to install the packages defined in the python/requirements.txt file using pip install -r requirements.txt. Typically, you want to create a new virtual environment before installing the packages, e.g. using pipenv shell. We ran the Python scripts using Python 3.12.
The SUND toolbox is built around a modular object-oriented approach, where models have outputs and inputs, which can come from other models or what we call "activities". In this project, the activities correspond to the drinks and meals.
Please note that we did not include simulations of the rejected alternative hypothesis in the python example, only the accepted model.
To run the Python version, either run the script in python/alcohol_main.py or the notebook python/alcohol_main.ipynb. We also put a version of the notebook at Google Colab, which can be used to run the notebook in the cloud.
The Python version of the model was used in a Streamlit application, to make it the model more generally available. The application is available here. The code for running the application is available at GitHub here, with a permanent copy of the code (as it was at the time of the submission/publication) available at Zenodo (DOI: 10.5281/zenodo.10054300).
All of the data used in the work is collected in the data.json file. Each entry in the JSON file corresponds to a single experiment (where we define a single experiments as giving a set of input and then measuring the outcome). This means that many experiments in our sense could come from the same original publication. For all entries in the JSON file, we have defined the inputs to our model, the measured data, and some additional meta information such as the DOI, and drink information.
For the inputs, we define the value of the input at t to the corresponding value of f. We use t = -Infinity to highlight what the value of an input would be before any stimulation starts.
All information not in the meta or input field, will be interpreted as an observable (such as ethanol concentration in the blood). An observable will contain Time, Mean and SEM fields with the corresponding values. If available, it could also contain individual measurement points in the Points field. An observable can also contain additional information, such as ylim for setting the limit of the y-axis when plotting the observable.
An entry can look like this:
"Mitchell_Beer": {
"meta": {
"doi": "10.1111/acer.12355",
"conc": 5.1,
"volume": 1.027,
"time": 20,
"kcal": 389.0
},
"input": {
"EtOH_conc": { "t": [-Infinity], "f": [5.1] },
"vol_drink_per_time": {
"t": [-Infinity, 0, 20],
"f": [0, 0.05134999999999999, 0]
},
"kcal_liquid_per_vol": {
"t": [-Infinity],
"f": [129.82]
},
"kcal_solid": { "t": [-Infinity], "f": [0] },
"sex": { "t": [-Infinity], "f": [1] },
"weight": { "t": [-Infinity], "f": [82.66] },
"height": { "t": [-Infinity], "f": [1.7711583490849163] }
},
"EtOH": {
"Mean": [
19.40065997, 34.59520339, 47.03908762, 45.56389434, 44.45473635,
42.79560442, 37.09797957, 31.58297765, 25.51879122, 20.00405244,
17.60258725
],
"SEM": [
7.875679573, 13.00359457, 11.17236552, 10.80606708, 6.04366109,
6.960196622, 4.945292065, 4.395581262, 4.578730481, 4.944765774,
4.395581262
],
"Time": [
19.748854001652564, 30.03962970175097, 44.73261792862444,
60.1758864053808, 75.27527643427419, 90.7526485587525,
119.78053671141895, 151.16441852754346, 180.5958665115863,
210.43371629764914, 241.05168702535147
],
"Unit": "mg/dL",
"ylim": [0, 100],
"xlim": [0, 250]
}
}If you use this model, please cite as:
Podéus H, Simonsson C, Nasr P, Ekstedt M, Kechagias S, Lundberg P, Lövfors W, Cedersund G (2024) A physiologically-based digital twin for alcohol consumption — predicting real-life drinking responses and long-term plasma PEth. npj Digital Medicine 7:112. https://doi.org/10.1038/s41746-024-01089-6
Or use the following bibtex entry:
@article{podeus_2024,
title = {A physiologically-based digital twin for alcohol consumption — predicting real-life drinking responses and long-term plasma {PEth}},
volume = {7},
issn = {2398-6352},
url = {https://doi.org/10.1038/s41746-024-01089-6},
doi = {10.1038/s41746-024-01089-6},
number = {1},
journal = {npj Digital Medicine},
author = {Podéus, Henrik and Simonsson, Christian and Nasr, Patrik and Ekstedt, Mattias and Kechagias, Stergios and Lundberg, Peter and Lövfors, William and Cedersund, Gunnar},
month = may,
year = {2024},
pages = {112},
}