Identification of flux checkpoints in a metabolic pathway through white-box, grey-box and black-box modeling approaches

Ophélie Lo-Thong, Philippe Charton, Xavier F. Cadet, Brigitte Grondin-Pérez, Emma Saavedra, Cédric Damour, Frédéric CadetScientific 2020 · DOI

TL;DR

This paper compares three modeling approaches (white-box, grey-box, and black-box) to understand the metabolic pathway of E. histolytica glycolysis. The grey-box approach, incorporating an adjustment term, emerged as the most robust, accurately predicting pathway flux and identifying key enzymes like PGAM and ENO as crucial for controlling the metabolic flow.

What Problem Does This Paper Address?

Metabolic pathway modeling is crucial for drug design and understanding biological systems, but it becomes incredibly challenging when experimental data is scarce or pathways are highly complex. Researchers often struggle to build accurate models that can predict how a pathway will behave under different conditions. This paper tackles this challenge by evaluating and comparing three distinct modeling strategies – white-box (detailed kinetic information), grey-box (kinetic with an adjustment), and black-box (artificial neural networks) – to model a specific part of the E. histolytica glycolysis pathway. The goal is to determine which approach is most effective and adaptable to varying levels of available data and pathway complexity.

Methods

The study employed three primary modeling methodologies:

  • White-box modeling: This approach utilizes detailed kinetic information, including enzyme parameters and rate equations, to build a mechanistic model. The researchers used COPASI, a complex pathway simulator, for this purpose.
  • Grey-box modeling: This hybrid approach builds upon the white-box model but introduces an "adjustment term" to the kinetic equation of a key enzyme (PPDK in this case). This term is not biologically derived but is empirically determined to improve model predictions, particularly when experimental data is limited or certain enzyme kinetics are poorly understood.
  • Black-box modeling: This method employs Artificial Neural Networks (ANNs) to learn the relationship between enzyme activities and pathway flux directly from experimental data, without explicit knowledge of the underlying biochemical mechanisms. The researchers experimented with different ANN architectures and activation functions.

All models were used to predict the final pathway flux and to determine flux control coefficients, which quantify the impact of each enzyme on the overall pathway flux.

Datasets

The study utilized experimental data from a previous publication (Moreno-Sanchez et al., 2008) for the second part of E. histolytica glycolysis. This data included enzyme activities and pathway flux measurements. The authors extracted this data from plots using WebPlotDigitizer. They used two sets of this data: one derived from experimental points and another from fitting curves. The COPASI model's predictions were also used to generate a test dataset for evaluating the generalization ability of the ANN models.

Key Insights and Ideas

  • The grey-box approach shines: The addition of an empirically derived adjustment term to a kinetic model significantly improved its predictive power, especially for enzymes with complex or poorly characterized kinetics (like PPDK). This highlights the value of hybrid approaches when pure mechanistic models fall short.
  • ANNs are fast but data-hungry: While ANNs offer rapid prediction capabilities, their performance is highly dependent on the quantity and quality of training data. They showed good predictive ability but struggled with generalization compared to the optimized grey-box model.
  • PGAM and ENO are the gatekeepers: Metabolic Control Analysis (MCA) consistently identified PGAM and, to a lesser extent, ENO as the primary enzymes controlling the flux through the second part of glycolysis. PPDK, despite being a key enzyme in the pathway, had a minimal impact on the overall flux.
  • Model choice matters: The study underscores that no single modeling approach is universally superior. The best choice depends on the specific problem, the available data, and the desired level of mechanistic understanding.

Key Results

  • All three modeling approaches (white-box, grey-box, and black-box) were capable of predicting the final pathway flux.
  • The grey-box model, incorporating an adjustment term for PPDK, demonstrated the best overall performance, achieving high R² values (up to 1.0) and low RMSE (as low as 0.23 nmol·min⁻¹) on the test set, indicating excellent predictive accuracy and generalization.
  • The ANN models showed good predictive ability on the training data but had limitations in generalization. The best ANN model achieved an R² of 0.98 on the test set.
  • Flux control analysis consistently identified PGAM as the most significant controlling enzyme (around 0.65-0.75), followed by ENO (around 0.11-0.21), with PPDK having a negligible impact (around 0.002-0.13).
  • The study found that while white-box models can be powerful, they often require extensive and accurate kinetic data, which is not always available.

Limitations and Future Work

  • Data dependency: The performance of all models, particularly ANNs, is heavily reliant on the quality and quantity of experimental data. The authors note that more data would improve ANN effectiveness.
  • PPDK kinetics: The study highlights that accurately modeling PPDK kinetics remains a challenge, even with the improved grey-box approach. Further research into its precise mechanism is needed.
  • Limited scope: The study focused on only the second part of glycolysis. The authors suggest merging this with the first part and other metabolic pathways to gain a more comprehensive understanding of E. histolytica metabolism.
  • Genetic regulation: Incorporating genetic-level regulations could further enhance model accuracy, but this requires experimental data on gene expression in the parasite.
  • Parameter estimation constraints: The authors suggest that future parameter estimation efforts should incorporate biologically relevant constraints to avoid unrealistic parameter values.

Summary updated

Abstract

Metabolic pathway modeling plays an increasing role in drug design by allowing better understanding of the underlying regulation and controlling networks in the metabolism of living organisms. However, despite rapid progress in this area, pathway modeling can become a real nightmare for researchers, notably when few experimental data are available or when the pathway is highly complex. Here, three different approaches were developed to model the second part of glycolysis of E. histolytica as an application example, and have succeeded in predicting the final pathway flux: one including detailed kinetic information (white-box), another with an added adjustment term (grey-box) and the last one using an artificial neural network method (black-box). Afterwards, each model was used for metabolic control analysis and flux control coefficient determination. The first two enzymes of this pathway are identified as the key enzymes playing a role in flux control. This study revealed the significance of the three methods for building suitable models adjusted to the available data in the field of metabolic pathway modeling, and could be useful to biologists and modelers.
BibTeX
@inproceedings{lothong2020identification,
  author = {Ophélie Lo-Thong and Philippe Charton and Xavier F. Cadet and Brigitte Grondin-Pérez and Emma Saavedra and Cédric Damour and Frédéric Cadet},
  title = {{Identification of flux checkpoints in a metabolic pathway through white-box, grey-box and black-box modeling approaches}},
  booktitle = {Scientific},
  year = {2020},
  doi = {10.1038/s41598-020-70295-5},
}
Figure