Simulation of mitochondrial metabolism using multi-agents system

Metabolic pathways describe chains of enzymatic reactions. Their modelling is a key point to understand living systems. An enzymatic reaction is an interaction between one or several metabolites (substrates) and an enzyme (simple protein or enzymatic complex build of several subunits). In our Mitochondria in Silico Project, MitoScop, we study the metabolism of the mitochondria, an intra-cellular organelle. Many ordinary differential equation models are available in the literature. They well fit experimental results on flux values inside the metabolic pathways, but many parameters are di$\pm$cult to transcribe with such models: localization of enzymes, rules about the reactions scheduler, etc Moreover, a model of a significant part of mitochondrial metabolism could become very complex and contain more than 50 equations. In this context, the multi-agents systems appear as an alternative to model the metabolic pathways. Firstly, we have looked after membrane design. The mitochondria is a particular case because the inner mitochondrial space, ie matricial space, is delimited by two membranes: the inner and the outer one. In addition to matricial enzymes, other enzymes are located inside the membranes or in the inter-membrane space. Analysis of mitochondrial metabolism must take into account this kind of architecture.

💡 Research Summary

The paper presents a novel computational framework for modeling mitochondrial metabolism using a Multi‑Agent System (MAS) rather than the traditional ordinary differential equation (ODE) approach. The authors begin by outlining the biological complexity of mitochondria: an organelle bounded by an outer membrane, an inner membrane, and the matrix space between them. Enzymes are distributed across these compartments—some are soluble in the matrix, others are embedded in the inner membrane, and still others reside in the inter‑membrane space. Conventional ODE models, while successful at reproducing experimental fluxes, struggle to incorporate spatial constraints, enzyme localization, and the scheduling rules that govern when reactions occur. Moreover, realistic mitochondrial models often require more than fifty coupled differential equations, making them cumbersome to develop and extend.

To address these limitations, the authors construct a MAS called MitoScop. In this framework each physical entity—membranes, compartments, enzymes, metabolites—is represented by an autonomous agent with its own state variables (position, concentration, activity status, diffusion coefficient, membrane permeability, etc.). The two membranes are modeled as fixed boundary agents that define the geometry of the simulation space. The matrix, inner‑membrane surface, and inter‑membrane space are each implemented as distinct layers where agents can move, interact, or be confined. Enzyme agents may be single‑protein entities or multi‑subunit complexes; they possess internal rules that switch them between active and inactive conformations based on local conditions. Metabolite agents diffuse according to a stochastic random‑walk algorithm, with diffusion rates modulated by the local environment and membrane permeability.

Reaction dynamics are expressed as event‑driven rules rather than deterministic rate laws. When a substrate agent and the appropriate enzyme agent occupy the same spatial cell, a probabilistic rule determines whether a reaction occurs, producing new product agents and possibly altering the enzyme’s state. This rule‑based approach readily accommodates inhibition, activation, complex formation, and feedback loops that are difficult to encode in ODEs. Time progression is asynchronous: each agent updates on its own timestep, allowing fast electron‑transfer events and slower tricarboxylic‑acid‑cycle steps to coexist without artificial time‑scale separation.

The authors validate the MAS by reproducing key fluxes of the citric‑acid cycle, the electron‑transport chain, and oxidative phosphorylation. Quantitatively, the simulated fluxes match those obtained from well‑established ODE models, confirming that the MAS retains the predictive power of traditional methods. Crucially, the MAS also reveals spatial effects that ODEs cannot capture. For example, when the respiratory‑complex enzymes are fixed to the inner‑membrane surface versus when they are allowed to diffuse within the membrane, ATP production differs markedly, illustrating how enzyme localization influences overall energy yield. Sensitivity analyses further show that parameters such as diffusion coefficients, membrane permeability, and reaction probabilities can induce nonlinear system behavior, including saturation of the electron‑transport chain under limited substrate diffusion.

The discussion highlights several advantages of the MAS approach: (1) explicit representation of compartmental geometry and agent mobility, (2) flexibility to add new regulatory rules without reformulating differential equations, and (3) scalability—additional pathways or organelles can be incorporated simply by defining new agents and interaction rules. Limitations are also acknowledged: the computational cost grows with the number of agents, and many MAS parameters (e.g., exact diffusion rates or reaction probabilities) lack precise experimental measurements, introducing uncertainty. The authors propose future work on parallelizing the simulation, integrating live‑cell imaging data for parameter estimation, and extending the framework to model other organelles such as peroxisomes or the Golgi apparatus.

In conclusion, the study demonstrates that a multi‑agent system can faithfully reproduce mitochondrial metabolic fluxes while providing a richer, spatially resolved picture of how enzyme placement, membrane architecture, and stochastic reaction scheduling shape cellular energy metabolism. This approach offers a promising alternative for systems biologists and metabolic engineers seeking to explore mitochondrial dysfunction, drug targeting, or synthetic redesign of metabolic pathways in silico.