FLUX-PREDICT

Implementation of the paper

"Combining multi-target regression deep neural networks and kinetic modeling to predict relative fluxes in reaction systems"

by Lucrezia Patruno, Francesco Craighero, Davide Maspero, Alex Graudenzi, Chiara Damiani

Requirements

All the results in the paper were obtained on Python 3.8.5 with the following requirements:

tensorflow==2.4.1
scikit_learn==0.24.2
pandas==1.2.3
numpy==1.19.5

Usage

To execute the script:

python grid_search.py

The outputs will be saved in the results folder.

Output

The script will save:

• the train/test partition as X_train.zip, y_train.zip, X_test.zip, y_test.zip.
• the sklearn.model_selection.GridSearchCV output in CVcomplete_results.pkl.
• the tensorflow best model in CVcomplete_best_estimator.h5.

To load the model and make predictions on the test set, perform the following steps (remember to change path_to_model and path_to_test):

import tensorflow as tf
import pandas as pd
from model import r2_metric

# Load model
model = tf.keras.models.load_model(
  "path_to_model/CVcomplete_best_estimator.h5", 
  custom_objects={"r2_metric": r2_metric})

# Load data
X_test = pd.read_csv("path_to_test/X_test.zip")
y_test = pd.read_csv("path_to_test/y_test.zip")

# Predict
y_pred = model.predict(X_test)

# Prediction to pandas dataframe
y_pred = pd.DataFrame(pred, columns=y_test.columns)

Data

To explore the data in the data folder, use the pandas library as follows:

import pandas as pd
pd.read_csv("data/X_full.zip")

Material

MetabolicModel.xml: SBML of yeast metabolic network model.