Modeling Metabolic Fluxes Using Deep Learning Based on Enzyme Variations: Applications to Glycolysis in Entamoeba histolytica

Metabolic pathway modeling, essential for understanding organism metabolism, is pivotal in predicting genetic mutation effects, drug design, and biofuel development. Enhancing these modeling techniques is crucial for achieving greater prediction accuracy and reliability. However, the limited experimental data or the complexity of the pathway makes it challenging for researchers to predict phenotypes. Deep learning (DL) is known to perform better than other Machine Learning (ML) approaches if the right conditions are met (i.e., a large database and good choice of parameters). Here, we use a knowledge-based model to massively generate synthetic data and extend a small initial dataset of experimental values. The main objective is to assess if DL can perform at least as well as other ML approaches in flux prediction, using 68,950 instances. Two processing methods are used to generate DL models: cross-validation and repeated holdout evaluation. DL models predict the metabolic fluxes with high precision and slightly outperform the best-known ML approach (the Cubist model) with a lower RMSE (≤0.01) in both cases. They also outperform the PLS model (RMSE ≥ 30). This study is the first to use DL to predict the overall flux of a metabolic pathway only from variations of enzyme concentrations.

Table of Contents

1. Prerequisites
2. Environment Setup
3. Running the Scripts
4. Project Structure
5. References

Prerequisites

Before installing this project, make sure you have the following installed on your machine:

• Python 3.8
• Conda
• Git (optional, but recommended)

Environment Setup

Using Conda, follow these steps to create the environment with all dependencies:

1. Clone the Git repository:

   git clone https://github.com/freddy-oulia/Metabolic_fluxes_DNN.git
   cd Metabolic_fluxes_DNN

2. Create a virtual environment:

   conda create -n environment_name python=3.8

3. Activate the virtual environment:

   conda activate environment_name

4. Install dependencies from requirements.txt:

   pip install -r requirements.txt

5. Deactivate the virtual environment:

   conda deactivate

6. Delete the virtual environment

   conda remove -n environment_name --all

Running the Scripts

This project includes several scripts, each designed for a specific task. Below are instructions for running each script.

1. Distribution

To see distribution of the dataset, please run the script located at show_distribution.py using this:

python scripts/show_distribution.py

2. PLS Model

The script for the PLS model is located at pls_model.py and is used to train and evaluate the PLS model. We can also make a figure to compare the expected values to the predicted values. The figure will be in plots/pls_model/pls_predicted_and observed.png. To run the code use this:

python scripts/pls_model.py

3. Cubist Model

The script for the Cubist model is located at cubist_model.py and is used to train and evaluate the Cubist model. We can also make a figure to compare the expected values to the predicted values. The figure will be in plots/cubist_model/cubist_predicted_and observed.png. To run the code use this:

python scripts/cubist_model.py

4. DNN Model with a simple holdout evaluation

The script for the DNN model is located at simple_run.py and is used to train and evaluate the DNN model following a holdout evaluation method. Results will be saved at results/simple_run/perf_simple_run.json and the figure with training and validation RMSE curves can be found at plots/simple_run/rmse_train_val.png. We can also make a figure to compare the expected values to the predicted values found at plots/simple_run/dnn_predicted_and observed.png In the hold_out_evaluation method we can change many hyperparameters such as the model architecture in the number_neurons variable, the loss function in the loss_fn variable, or even the Earlystopping function in the add_earlystopping boolean variable.

python scripts/simple_run.py

5. DNN Model with repeated holdout evaluation

The script to run the repeated hold out evaluation method with a DNN is found at repeated_hold_out_evaluation_run.py. Results are saved in results/rho/performances_rho.json. To make the boxplots or the correlation metrics plots change the make_boxplot boolean or the correlation_metrics boolean to True and make sure that the results file is in the right location. Plots can be found in plots/boxplots and plots/correlation.

python scripts/repeated_hold_out_evaluation_run.py

6. DNN Model with repeated cross-validation

The script to run the repeated cross-validation method with a DNN is found at cross_validation_run.py. Results are saved in results/cv/performances_cv.json. To make the boxplots or the correlation metrics plots change the make_boxplot boolean or the correlation_metrics boolean to True and make sure that the results file is in the right location. Plots can be found in plots/boxplots and plots/correlation.

python scripts/cross_validation_run.py

7. Boxplots comparison

The script to make the boxplots comparison is located at boxplot_comparison.py. The generated figures are saved in plots/boxplot/comparison/. Please make sure that the results file for the repeated cross-validation and repeated hold out evaluation are correctly located.

python scripts/boxplot_comparison.py

Project structure

Here’s an overview of the project structure to help you navigate between the various files and folders:

Metabolic_fluxes_DNN/
│
├── data/                   # Contains datasets
│   ├── Table_S1_training_set.csv
│   └── Table_S2_test_set.csv
│
├── results/                # Generated  results
│   ├── simple_run
│       └── perf_simple_run.json
│   ├── rho
│       └── performances_rho.json
│   └── cv
│       └── performances_cv.json
│
│
├── plots/                # Generated figures
│   ├── distribution
│       ├── ...
│       └── distribution_Training_set_pgam.png
│   ├── pls_model
│       └── pls_predicted_and observed.png
│   ├── cubist_model
│       └── cubist_predicted_and observed.png
│   ├── simple_run
│       ├── rmse_train_val.png
│       └── dnn_predicted_and observed.png
│   ├── boxplot
│       ├── cross-validation
│            ├── rmse.png
│            ├── mae.png
│            └── r2.png
│       └── repeated_hold_out_evaluation
│            ├── rmse.png
│            ├── mae.png
│            └── r2.png
│   ├── correlation
│       ├── Kendall
│            ├── rmse_cv.png
│            ├── ...
│            └── r2_rho.png
│       ├── Spearman
│            ├── rmse_cv.png
│            ├── ...
│            └── r2_rho.png
│       └── Pearson
│            ├── rmse_cv.png
│            ├── ...
│            └── r2_rho.png
│   └── boxplot_comparison
│       ├── rmse.png
│       ├── mae.png
│       └── r2.png
│
│   # Scripts
├── show_distribution.py
├── pls_model.py
├── cubist_model.py
├── simple_run.py
├── repeated_hold_out_evaluation_run.py
├── cross_validation_run.py
├── boxplot_comparison.py
├── create_dnn_model.py
├── gridearch_run.py
├── metrics.py
├── results_plots.py
│
├── requirements.txt        # Python dependencies for pip
├── README.md               # Project documentation
├── LICENSE-CC-BY.md        # Data license
└── LICENSE                 # Project license

References

Oulia, F.; Charton, P.; Lo-Thong-Viramoutou, O.; Acevedo-Rocha, C.G.; Liu, W.; Huynh, D.; Damour, C.; Wang, J.; Cadet, F. Metabolic Fluxes Using Deep Learning Based on Enzyme Variations: Application to Glycolysis in Entamoeba histolytica. Int. J. Mol. Sci. 2024, 25, 13390. https://doi.org/10.3390/ijms252413390

License

The complete source code, datasets, and models are available under the Creative Commons Attribution-Non-Commercial ShareAlike International License, Version 4.0 (CC-BY-NC-SA 4.0) for open, non-commercial use. See the LICENSE-CC-BY-NC-SA file for more details.