MDFM: Microbiome Data Fusion Model

Table of Contents

1. About The Project
2. Theoretical Background
   • Data Fusion Model
3. Getting Started
   • Prerequisites
   • Installation
4. Usage
• Download Data
• Model Calibration
• Model Prediction

About The Project

fusion_model is a tool developed for the project Zukunftslabor2030 (ZL2030) that aims the dynamic assessment and prediction of food spoilage on the basis of different measurement data. This tool focuses on developing a temperature-dependent data fusion shelf-life model by analyzing bacterial count data (MiBi) and bacterial diversity data (NGS, MALDI-ToF). By calibrating the model with microbiological measurements, it predicts the shelf life of the product based on input storage temperature conditions and an initial number of bacteria.

(back to top)

Theoretical Background

Data Fusion Model

The mathematical model developed for data fusion of the microbiome data could be soon found in the coming out paper.

(back to top)

Getting Started

Prerequisites

For this tool a working installation of the python scripting language version python 3.11 is required. All necessary libraries could be found in requirements.txt and can be installed thin pip:

  pip install -r requirements.txt

Installation

To get started with the fusion_model project the user is expected to clone the git repository:

   git clone https://github.com/polinagaindrik/fusion_model

(back to top)

Usage

To start with the Digital Twin project we use a main.py file that allows us to download data, calibrate the model, and predict the behavior of the model for input temperature conditions. Below is a brief overview of the primary functions available.

At the beginning we import the functions of the source code to our main file:

  import os
  import sys
  sys.path.append(os.getcwd())
  import DT_diversity as fm

(back to top)

Download Data

The Excel sheets containing microbiology (MiBi) experimental data stored in OpenBis can be easily downloaded with the function:

fm.data.download_microbiology_from_openbis(path, username, password)

Here path="/../" defines the location where Excel files are saved. username, password are personal login data for OpenBis.

MALDI-ToF data:

fm.data.download_MALDI_from_openbis(path, username, password)

NGS data:

fm.data.download_NGS_from_openbis(path, username, password)

To prepare the ZL2030 data for model calibration use the function:

dfs_calibr, bact, T, s_predef = fm.data.prepare_ZL2030_data(exps_mibi, exps_maldi, exps_ngs, cutoff=0.01)

where exps_mibi, exps_maldi, exps_ngs define available experiments for each data type that we want to use for model calibration.

Or another way to get the data is to generate the in-silico data for a defined insilico_model (e.g. from fm.data.):

dfs_calibr, bact, T, s_predef =  fm.data.prepare_insilico_data(insilico_model, temps, ntr)

(back to top)

Model Calibration

At the beginning we need to define calibration setup, for example:

calibr_setup={
        'model': ode_model,
        'param_bnds':p_bnds,
        'T_x': T,
        'workers': -1,
        'output_path': path,
        'n_cl': n_cl,
        'n_media': n_media,
        'dfs': dfs_calibr,
        'aggregation_func': fm.pest.cost_sum_and_geometric_mean,
        'exps': exps_mibi,
        'exp_temps': ['V01':2.,]
        }

To run the parameter estimation function:

param_opt = fm.pest.calculate_model_params(fm.pest.cost_withS, calibr_setup)[0]

To plot the estimated model for each experiment:

fm.plotting.plot_optimization_result(param_opt, calibr_setup, time_array)

(back to top)

Model Prediction

The initial conditions for each experiment are different and needed to be estimated for correct model prediction. For this we define prediction setup with param_ode and s_x estimated in param_opt and dfs_predict, exps_predict defining experiments for prediction:

    prediction_setup = calibr_setup
    prediction_setup['param_ode'] = param_ode
    prediction_setup['dfs'] = dfs_predict
    prediction_setup['param_bnds'] = x0_bnds
    prediction_setup['exps'] = exps_predict
    prediction_setup['s_x'] = s_x

To run the initial values estimation for the model prediction:

x0_opt = fm.pest.calculate_prediction(fm.pest.cost_initvals, prediction_setup)[0]

To plot the estimated model for each experiment:

fm.plotting.plot_prediction_result(prediction_setup['param_ode'], x0_opt, prediction_setup, time_array)