Metabolic Fluxes Using Deep Learning Based on Enzyme Variations: Application to Glycolysis in Entamoeba histolytica.

Area of Science:

• Metabolic pathway modeling
• Systems biology
• Computational biology

Background:

• Metabolic pathway modeling is crucial for predicting effects of genetic mutations, drug design, and biofuel development.
• Challenges in phenotype prediction arise from limited experimental data and pathway complexity.
• Deep learning (DL) excels with large datasets and optimal parameters.

Purpose of the Study:

• To assess if DL models can achieve comparable or superior performance to other machine learning (ML) approaches in metabolic flux prediction.
• To leverage a knowledge-based model for generating synthetic data to augment limited experimental datasets.
• To evaluate DL performance using cross-validation and repeated holdout methods.

Main Methods:

• Utilized a knowledge-based model to generate a large synthetic dataset (68,950 instances) from a small experimental dataset.
• Developed and evaluated Deep Learning (DL) models using cross-validation and repeated holdout.
• Compared DL model performance against established ML models like Cubist and PLS.

Main Results:

• DL models demonstrated high precision in predicting metabolic fluxes.
• DL models slightly outperformed the Cubist model, achieving a lower Root Mean Square Error (RMSE) (≤0.01).
• DL models significantly outperformed the PLS model (RMSE ≥ 30).

Conclusions:

• Deep learning models can effectively predict metabolic pathway fluxes using only enzyme concentration variations.
• This study is the first to apply DL for predicting overall metabolic pathway flux from enzyme concentration data.
• DL offers a promising approach for enhancing the accuracy and reliability of metabolic pathway modeling.