dAMN — Hybrid Neural Network for Dynamical FBA

First usable release of the dAMN software.
This version corresponds to the implementation used to generate results for the associated publication.

Overview

dAMN is a hybrid machine learning model that combines:

• A neural network for metabolic flux inference;
• A dynamical FBA to simulate metabolite and biomass evolution over time

It is designed to predict time-course biomass growth under various media conditions, integrating stoichiometry and transport constraints from genome-scale metabolic models (GEMs).

Applied on E.coli dataset.

---

Project Structure

File/Folder	Description
dAMN.ipynb	Notebook to train, test and parametrize the model with a given dataset
dAMN_train.py	Script to train the model on a given dataset
dAMN_test.ipynb	Notebook to test and visualize the prediction
dAMN_parameter_search.py	Script for model parametrization
utils.py	Core functions for data preprocessing, model training, testing and plotting
data/	Input datasets: media, OD, and metabolic model (SBML)
model/	Folder where trained models and validation arrays are stored
figure/	Plots for training and testing curves
environment.yml	conda environment configuration file

---

Setup and Environment

This project uses Python ≥ 3.8, TensorFlow 2.19.0, and COBRApy.

Create the Conda environment (recommended)

To recreate the required environment from the environment.yml file:

conda env create -n dAMN_env -f environment.yml
conda activate dAMN_env

Contributors

Pr. Jean-Loup Faulon (jean-loup.faulon@inrae.fr): conceptualization, coding, modeling

Danilo Dursoniah, Postdoc (danilo.dursoniah@inrae.fr): testing, maintenance