Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Linking intra- and extra-cellular metabolic domains via neural-network surrogates for dynamic metabolic control

Sebastián Espinel-Ríos, José L. Avalos

TL;DR

This work proposes a solution involving a machine-learning surrogate derived from steady-state constraint-based metabolic modeling that bridges the gap between manipulatable intracellular fluxes and process exchange rates and develops hybrid machine- learning-supported dynamic models.

Abstract

We outline a modeling and optimization strategy for investigating dynamic metabolic engineering interventions. Our framework is particularly useful at the early stages of research and development, often constrained by limited knowledge and experimental data. Elucidating a priori optimal trajectories of manipulatable intracellular fluxes can guide the design of suitable control schemes, e.g., cyber(ge)netic or in-cell approaches, and the selection of appropriate actuators, e.g., at the transcriptional or post-translational levels. Model-based dynamic optimization is proposed to predict optimal trajectories of target manipulatable intracellular fluxes. A challenge emerges as existing models are often oversimplified, lacking insights into metabolism, or excessively complex, making them difficult to build and implement. Here, we use surrogates derived from steady-state solutions of constraint-based metabolic models to link manipulatable intracellular fluxes to the process exchange rates of structurally simple hybrid dynamic models. The latter can be conveniently used in optimal control problems of metabolism. As a proof of concept, we apply our method to a reduced metabolic network of $\textit{Escherichia coli}$ considering two different scenarios of dynamic metabolic engineering.

Linking intra- and extra-cellular metabolic domains via neural-network surrogates for dynamic metabolic control

TL;DR

This work proposes a solution involving a machine-learning surrogate derived from steady-state constraint-based metabolic modeling that bridges the gap between manipulatable intracellular fluxes and process exchange rates and develops hybrid machine- learning-supported dynamic models.

Abstract

We outline a modeling and optimization strategy for investigating dynamic metabolic engineering interventions. Our framework is particularly useful at the early stages of research and development, often constrained by limited knowledge and experimental data. Elucidating a priori optimal trajectories of manipulatable intracellular fluxes can guide the design of suitable control schemes, e.g., cyber(ge)netic or in-cell approaches, and the selection of appropriate actuators, e.g., at the transcriptional or post-translational levels. Model-based dynamic optimization is proposed to predict optimal trajectories of target manipulatable intracellular fluxes. A challenge emerges as existing models are often oversimplified, lacking insights into metabolism, or excessively complex, making them difficult to build and implement. Here, we use surrogates derived from steady-state solutions of constraint-based metabolic models to link manipulatable intracellular fluxes to the process exchange rates of structurally simple hybrid dynamic models. The latter can be conveniently used in optimal control problems of metabolism. As a proof of concept, we apply our method to a reduced metabolic network of considering two different scenarios of dynamic metabolic engineering.
Paper Structure (12 sections, 4 equations, 6 figures)

This paper contains 12 sections, 4 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Hybrid modeling strategy
  3. Constraint-based model in steady state
  4. Mapping intra- and extra-cellular metabolic fluxes
  5. Hybrid dynamic model linked to intracellular fluxes
  6. Dynamic metabolic control
  7. Dynamic optimization
  8. Proof of concept
  9. Scenario 1
  10. Scenario 2
  11. Scenario 3
  12. Conclusion

Figures (6)

  • Figure 1: Example of an FBA simulation for the considered E. coli's metabolic network. FBA simulation and network map obtained with CNApy thiele_cnapy_2022. The shown scenario belongs to a wild-type flux distribution of E. coli under anaerobic conditions.
  • Figure 2: Explored flux space used to build the surrogate model in scenario 1.
  • Figure 3: Selected parity plots of the trained neural-network surrogate of FBA (scenario 1) for the (a) biomass and (b) ethanol extracellular fluxes.
  • Figure 4: Effect of (a) a given trajectory of $V_\mathrm{ack}$ and (b) the predicted dynamic profiles of relevant extracellular states in scenario 1.
  • Figure 5: Effect of (a) an optimized trajectory of $V_\mathrm{ack}$ and (b) the predicted dynamic profiles of relevant extracellular states in scenario 2.
  • ...and 1 more figures