A Machine Learning Approach for Efficient Selection of Enzyme Concentrations and Its Application for Flux Optimization

Anamya Ajjolli Nagaraja, Philippe Charton, Xavier F. Cadet, Nicolas Fontaine, Mathieu Delsaut, Birgit Wiltschi, Alena Voit, Bernard Offmann, Cédric Damour, Brigitte Grondin-Pérez, Frédéric CadetCatalysts 2020 · DOI

TL;DR

This paper introduces a novel "glass ceiling" artificial neural network (GC-ANN) approach to overcome the limitations of standard ANNs in predicting optimal enzyme concentrations for maximum flux in cell-free systems. By combining PCA, decision trees, and ANNs, the method successfully identified enzyme balances that yield significantly higher fluxes (up to 63% improvement) and reduced costs (up to 25%) compared to existing methods.

What Problem Does This Paper Address?

Metabolic engineering aims to optimize biological pathways for producing valuable molecules. Cell-free systems offer advantages over engineered microbes, but a key challenge is determining the optimal concentrations of multiple enzymes to maximize product yield. Traditional methods for this optimization are often expensive and time-consuming. Artificial Neural Networks (ANNs) can predict flux based on enzyme concentrations, but they struggle to extrapolate beyond their training data, creating a "glass ceiling" where predicted fluxes plateau, even if higher actual fluxes are achievable. This paper addresses the need for a method that can effectively explore and predict these "out-of-the-box" optimal enzyme concentrations, leading to improved yields and cost-efficiency in cell-free bioproduction.

Methods

The core of this work is the development of the Glass Ceiling ANN (GC-ANN) methodology. It involves three main stages:

  1. Preparation:
    • Principal Component Analysis (PCA): Used for dimensionality reduction to identify correlations between variables, though its impact was limited in this specific study due to constant total enzyme concentration.
  2. Data Classification (Decision Trees & Discriminant Analysis): Used to derive a rule for identifying enzyme concentration combinations that lead to high flux values (specifically, > 12 µM/s).
  3. ANN Model: A standard artificial neural network was trained to predict flux based on enzyme concentrations.
  4. Execution:
    • New Enzyme Concentration Generation: New combinations of enzyme concentrations were generated within a defined range.
  5. Flux Prediction: The trained ANN was used to predict fluxes for these new concentrations. The GC-ANN approach specifically focused on predicted fluxes that hit the "glass ceiling" (around 12 µM/s) and also satisfied the derived high-flux rule.
  6. Validation:
    • In Silico Validation: A kinetic model of the upper glycolysis pathway was built and used to simulate fluxes for the selected enzyme balances. This served as a computational validation of the GC-ANN predictions.
  7. Experimental Validation: The most promising enzyme balances identified by GC-ANN were tested in vitro to experimentally determine their flux and cost-efficiency.

Datasets

The primary dataset used in this study was derived from Fievet et al. [43]. This dataset comprised 121 experimental balances of four enzymes (PGI, PFK, FBA, and TPI) in the upper part of glycolysis, along with their corresponding flux values. The total enzyme concentration across these four enzymes was kept constant at 101.9 mg/L. This dataset was used for training the ANN model and deriving the high-flux rule.

Key Insights and Ideas

  • The "Glass Ceiling" is Real and Explorable: The study confirms that standard ANNs indeed hit a prediction ceiling. However, the crucial insight is that this ceiling doesn't represent the absolute maximum achievable flux; higher fluxes can exist beyond the ANN's direct prediction range.
  • Combining Statistical Methods with ANNs is Powerful: The GC-ANN approach demonstrates that integrating data classification (decision trees) with ANNs can effectively guide the search for optimal conditions that lie outside the typical prediction space.
  • High Flux Doesn't Necessarily Mean High Cost: The research highlights that achieving higher metabolic flux doesn't automatically translate to prohibitive costs. The GC-ANN methodology allows for the identification of enzyme balances that offer both high flux and cost-efficiency.
  • Experimental Validation is Crucial: While in silico methods are valuable for initial screening, the study emphasizes the necessity of experimental validation to confirm the "out-of-the-box" improvements predicted by the GC-ANN approach.

Key Results

  • Flux Improvement: The GC-ANN methodology successfully identified enzyme balances that resulted in flux improvements of up to 63% compared to the original experimental data.
  • Cost Reduction: These flux improvements were coupled with a significant cost decrease, with reductions of up to 25% observed for the assay.
  • Kinetic Model Validation: The kinetic model built for the upper glycolysis pathway showed good agreement with experimental data (R² = 0.84 and R² = 0.99 for different validation sets), validating its quality for in silico predictions.
  • ANN Performance: The ANN model achieved an RMSE of 0.84 and an R² of 0.93 in previous work [44] and was used here for flux prediction.
  • Experimental Control: The control experiments using Fievet et al.'s data showed a high correlation (R² = 0.99) between experimental and kinetic model-predicted fluxes.

Limitations and Future Work

  • Dataset Dependence: The performance of the GC-ANN approach, like any machine learning method, is dependent on the quality and representativeness of the training dataset.
  • Generalizability: While the methodology was demonstrated for the upper part of glycolysis, its direct applicability to other pathways would require new datasets and potentially adjustments to the statistical methods.
  • Future Work: The authors suggest exploring the use of the validated kinetic model to generate more data for the ANN-based model, further expanding the exploration of enzyme balances. They also aim to apply this methodology to other pathways and are actively seeking new experimental datasets for this purpose. The potential for rapid optimization of cell-free systems in industrial environments is a key future direction.

Summary updated

Abstract

The metabolic engineering of pathways has been used extensively to produce molecules of interest on an industrial scale. Methods like gene regulation or substrate channeling helped to improve the desired product yield. Cell-free systems are used to overcome the weaknesses of engineered strains. One of the challenges in a cell-free system is selecting the optimized enzyme concentration for optimal yield. Here, a machine learning approach is used to select the enzyme concentration for the upper part of glycolysis. The artificial neural network approach (ANN) is known to be inefficient in extrapolating predictions outside the box: high predicted values will bump into a sort of “glass ceiling”. In order to explore this “glass ceiling” space, we developed a new methodology named glass ceiling ANN (GC-ANN). Principal component analysis (PCA) and data classification methods are used to derive a rule for a high flux, and ANN to predict the flux through the pathway using the input data of 121 balances of four enzymes in the upper part of glycolysis. The outcomes of this study are i. in silico selection of optimum enzyme concentrations for a maximum flux through the pathway and ii. experimental in vitro validation of the “out-of-the-box” fluxes predicted using this new approach. Surprisingly, flux improvements of up to 63% were obtained. Gratifyingly, these improvements are coupled with a cost decrease of up to 25% for the assay.
BibTeX
@inproceedings{nagaraja2020machine,
  author = {Anamya Ajjolli Nagaraja and Philippe Charton and Xavier F. Cadet and Nicolas Fontaine and Mathieu Delsaut and Birgit Wiltschi and Alena Voit and Bernard Offmann and Cédric Damour and Brigitte Grondin-Pérez and Frédéric Cadet},
  title = {{A Machine Learning Approach for Efficient Selection of Enzyme Concentrations and Its Application for Flux Optimization}},
  booktitle = {Catalysts},
  year = {2020},
  doi = {10.3390/catal10030291},
}
Figure