Ecology’s Energy Language Rewrites How Immune Memory Is Modeled

…

The publication of this work arrives at a moment when the limitations of reductionist pipelines in biomedicine are becoming increasingly visible. High-throughput technologies have produced enormous catalogs of molecular interactions, yet translating those catalogs into predictions about how a whole organism responds to a perturbation remains stubbornly difficult. The authors argue that part of the difficulty is architectural: most computational approaches assemble their picture of a system from very large numbers of local components, and the systemic character of the dynamics—feedback, self-organization, regime shifts—has to be inferred afterward rather than built in from the start. Energy Systems Language inverts that logic. It begins with a small set of state variables chosen because they carry the flows of matter, energy, and information that define the system, and it encodes the feedback structure directly in the diagram before any equation is written.

The intellectual lineage of the approach is worth emphasizing. The symbolic notation descends from the system dynamics tradition initiated by Jay Forrester at MIT, but its biological form was shaped decisively by Howard T. Odum, who spent decades developing a rigorous algebraic framework grounded in the thermodynamics of open systems. Odum’s central insight was that very different kinds of flows—solar radiation, water, biomass, money, information—could be placed on a common footing by expressing them in energy terms, allowing heterogeneous parts of a system to be compared and balanced within a single accounting scheme. In ecology this produced models of ecosystems, estuaries, and whole economies of nature that could be simulated over decades. What the present work demonstrates is that the same machinery, with an ATP-equivalent unit serving as the common currency, can be pointed at the machinery of the vertebrate immune system without any fundamental modification of the formalism.

That transfer is less trivial than it may appear. In ecological applications, the stocks are typically pools of biomass or nutrients, and the processes are feeding, respiration, and recycling. In the immune model, the stocks are populations of cells—naive and activated lymphocytes, germinal-center participants, plasma cells of varying lifespans—and the flows are proliferation, differentiation, death, and antibody secretion. The conceptual bridge is that every one of these biological processes has an energetic cost that can be expressed in the shared unit, and every regulatory interaction can be drawn as a flow of information or control. The authors are careful to distinguish real flows, which physically move matter or energy between stocks and therefore enter the balance equations, from control flows, which regulate processes but carry negligible energy of their own and are deliberately excluded from the outflow accounting of a stock. This distinction, borrowed from engineering practice, prevents the common modeling error of double-counting regulatory signals as material resources.

The choice of the antigen-specific adaptive immune response as a proof of concept is strategically astute. Humoral immunity is one of the best-characterized dynamic systems in biomedicine, with a rich experimental literature documenting germinal-center kinetics, the lifespan distributions of plasma cell subsets, and the decay of circulating antibody over years. That empirical richness matters enormously for a modeling framework in which parameters are constrained by observation rather than fitted freely. The simulator reproduces phenomena operating on strikingly different time scales simultaneously: the days-long cycles of germinal-center reaction, the monthly establishment of durable antibody-secreting populations, and the decadal slow decline of antibody titers in the absence of re-exposure. Capturing all of these within one coherent set of coupled differential equations is a demanding test, because errors in the fast dynamics propagate into the slow ones, and the model must conserve its accounting across all of them.

The thermodynamic grounding deserves particular attention because it is what separates this approach from purely phenomenological curve-fitting. Open systems thermodynamics holds that living systems maintain their organization by continuously dissipating energy gradients, and that the structure of those gradients—the sources, sinks, storage, and recycling pathways—constrains what dynamic behaviors are physically possible. By building models whose bookkeeping respects those constraints, ESL simulators inherit a form of physical plausibility: they cannot invent unlimited growth from finite resources, and they cannot sustain oscillations without an energy source. In biomedical terms, this means that a model of a lymphoid compartment cannot produce more plasma cells than the available activation events and proliferation capacity permit, a discipline that purely statistical models of cell populations often lack.

The question of spatial constraints, which the authors single out as a persistent problem in cell-population modeling, illustrates the value of treating space as a stock. Immune memory formation and plasma-cell malignancies both depend critically on where cells reside: germinal centers are physically bounded structures within lymphoid tissue, and the survival of long-lived plasma cells depends on limited niches in the bone marrow. Conventional compartment models often handle this with ad hoc carrying-capacity terms whose biological justification is unclear. In the ESL framework, units of space can be represented as a quasi-energetic stock that is consumed and released by the processes occupying it, making crowding, competition for niches, and the saturation of lymphoid architecture explicit components of the dynamics rather than afterthoughts. This is a genuinely useful formal move, and it is the kind of thing that becomes natural only when the modeling language treats extensive quantities uniformly.

The contrast drawn with network analysis and related data-driven methods should be read as complementary rather than adversarial. Network medicine has been extraordinarily successful at identifying hubs, modules, and topological vulnerabilities in molecular interaction graphs, and those findings supply exactly the biomedical knowledge from which ESL diagrams can be constructed. The two approaches operate at different levels of description: one characterizes the static wiring of a system, the other the dynamic behavior that wiring produces over time. A plausible division of labor emerges in which network methods inform which stocks and interactions belong in a diagram, while the ESL simulator tests whether the proposed structure actually generates the observed temporal patterns. Discrepancies between simulated and observed dynamics then become diagnostic, pointing to missing feedbacks or misestimated flows rather than simply degrading a fit statistic.

The emphasis on what-if exploration connects this work to a broader aspiration in medicine: the construction of disease digital twins. The idea of a patient-specific computational surrogate that can be interrogated before committing to a therapeutic intervention has gained traction across oncology and immunology, but most implementations remain either too abstract to be mechanistic or too mechanistic to be tractable. An ESL-based simulator occupies an interesting middle ground. Because its parameters are tied to empirical observation and biophysical constraints, it can in principle be personalized—adjusted to reflect a particular patient’s measured cell counts, antibody levels, or tissue capacities—without abandoning its mechanistic structure. Because it is dynamic, it can be run forward under hypothetical interventions, asking how a proposed drug schedule, cell-depletion regimen, or antigen re-exposure would alter the trajectory of the system.

The application to lymphoid malignancies and autoimmune diseases named by the authors is well chosen, because both classes of pathology are paradigmatically systemic. Autoimmune disease involves feedback loops in which tissue damage releases antigens that further stimulate the immune response, creating self-amplifying cycles that no single molecular target fully explains. Lymphoid malignancies arise from the same cell populations the model represents—germinal-center B cells and their descendants—and their clinical behavior, including dormancy, relapse, and transformation, exhibits the regime shifts and patient-to-patient variability that linear models handle poorly. A simulator that captures the normal dynamics of these compartments provides a baseline against which pathological rewiring can be expressed as specific, testable modifications to the diagram: a broken feedback, an altered carrying capacity, a redirected flow.

There are, of course, challenges that the proof of concept does not yet resolve. Selecting the right state variables requires judgment, and different modelers might draw different diagrams for the same biological system, with consequences that are not always easy to anticipate. Parameter estimation for processes operating across decades, such as plasma-cell longevity, depends on sparse and indirect measurements. Validation of what-if predictions ultimately requires experiments or clinical observations that the model itself cannot replace. And the translation from a symbolic diagram to a system of coupled differential equations, while systematic, still involves choices about functional forms for interactions that can shape simulated behavior. These are the ordinary difficulties of mechanistic modeling, but they are worth stating plainly, because the framework’s value will be judged by how well its predictions survive contact with data.

What the work ultimately offers is a demonstration that a modeling language developed for ecological complexity, with roots stretching back to mid-twentieth-century systems thinking, can be brought to bear on one of the most intricate dynamic systems in human biology without losing its rigor. The immune system, with its layered feedbacks, its memory, and its capacity for self-organization across time scales, is a demanding test case, and the successful reproduction of its hallmark behaviors suggests that the formalism has genuine reach. If subsequent studies extend the approach to pathological variants of the same system, and if patient-specific parameterization proves feasible in practice, the vision of computational tools that support hypothesis testing for individualized therapeutic strategies will have moved from aspiration toward method. The deeper contribution may be conceptual: a reminder that the dynamics of disease are properties of organized systems, and that a language designed to express organization—stocks, flows, feedback, and the thermodynamic constraints that bind them—may be exactly what biomedicine needs to reason clearly about complexity.

Subject of Research: Thermodynamics-based energy systems modeling of the adaptive immune response

Article Title: Addressing biomedical complexity using the energy systems language

Article References: Addressing biomedical complexity using the energy systems language. (n.d.). https://doi.org/10.3389/fsysb.2026.1903662

Image Credits: AI Generated

DOI: 10.3389/fsysb.2026.1903662

Keywords: energy systems language, systems biology, immune response, plasma cells, antibodies, thermodynamics, computational modeling, digital twin, multiple myeloma, autoimmune disease, bone marrow niche, feedback loops