From Hallmarks to Control Laws: A Control-Theoretic Framework for Aging Drug Discovery

§Abstract

Aging research has produced powerful explanatory and classificatory frameworks, including evolutionary theories, damage and maintenance theories, the Hallmarks of Aging, SENS, geroscience, hyperfunction theory, and information-loss models. These frameworks have substantially clarified why late-life decline can emerge, what biological processes change with age, and which classes of damage or dysregulation may be therapeutically relevant. Yet drug discovery requires a further layer of theory: a quantitative rule for determining which intervention, in which biological state, at which dose, at what time, and in what sequence, will safely restore or maintain organismal function. Current frameworks generally characterize what changes during aging, but they do not by themselves define equations of motion, intervention-response operators, safety constraints, optimality conditions, or falsifiable rules for target prioritization and combination design.

Here we propose a control-theoretic framework for aging drug discovery. We model a tissue, organ, or organism as a biological state x(t) evolving under endogenous dynamics, stochastic perturbations, and admissible interventions. Aging is defined as progressive loss of safe controllability: the increasing cost and decreasing feasibility of returning a biological system to a functional viability set under safety-constrained interventions. In this formulation, biological age is not merely a correlation with chronological age or a deviation from youth, but the minimum safe control cost required to restore or maintain function. Formally, aging corresponds to an increase in the optimal control value function V(x₀, T), where

subject to controlled stochastic biological dynamics. In this framework, drugs are vector fields on biological state space, targets are ranked by their expected reduction of restoration cost, and combination therapies are evaluated by their ability to expand the reachable safe set.

We further describe how multi-model AI orchestration, perturbational omics, longitudinal biomarkers, single-cell atlases, clinical response data, and AI drug discovery pipelines can be integrated to estimate control-relevant quantities. As a validation strategy, we outline retrospective scoring of 30 AI-discovered preclinical candidate compounds according to Hallmark annotation, age association, disease relevance, network centrality, and control-value reduction, with the hypothesis that control-value models better predict translational advancement and toxicity-adjusted efficacy than Hallmark membership alone. To make the formalism concrete, we provide a worked five-dimensional example of aged murine liver in which a senolytic-first-then-reprogramming protocol outperforms the reverse order; the effect is traced explicitly to a non-vanishing Lie bracket between the senolytic and reprogramming vector fields. We then situate this framework relative to network controllability theory and classical systems pharmacology, and we formally address structural and practical identifiability, model misspecification, and robust control as prerequisites for translation. The framework generates experimentally distinguishable predictions, including state-dependent drug efficacy, sequence-dependent intervention effects, reversibility boundaries for reprogramming, high-value non-Hallmark targets, and irreversible loss of controllability in structurally aged tissues. This framework does not replace existing theories of aging; it provides an interventional layer designed to make aging theory operational for therapeutic discovery.

1.Introduction: From Mechanisms to Intervention Laws

Aging biology has entered a period of extraordinary conceptual and technological expansion. Evolutionary theories explain why natural selection may permit late-life functional decline (Medawar, 1952; Williams, 1957; Hamilton, 1966). Damage and maintenance theories describe the accumulation of molecular and cellular lesions over time (Harman, 1956; Kirkwood, 1977). The Hallmarks of Aging organize diverse mechanisms into a widely used ontology of genomic instability, telomere attrition, epigenetic alteration, loss of proteostasis, deregulated nutrient sensing, mitochondrial dysfunction, cellular senescence, stem-cell exhaustion, altered intercellular communication, disabled macroautophagy, chronic inflammation, and dysbiosis (López-Otín et al., 2013; López-Otín et al., 2023). SENS provides a repair-oriented catalogue of damage classes and candidate interventions (de Grey et al., 2002; de Grey, 2007). Information theories emphasize epigenetic drift, loss of cellular identity, and partial reversibility through reprogramming (Lu et al., 2020; Sinclair and LaPlante, 2019; Yang et al., 2023). Geroscience frames aging mechanisms as modifiable drivers of chronic disease and seeks interventions that delay or prevent multiple age-related pathologies (Kennedy et al., 2014; Sierra and Kohanski, 2017; Barzilai et al., 2018). Hyperfunction theory highlights persistent growth signaling, especially mTOR-dependent programs, as drivers of late-life pathology (Blagosklonny, 2006; Blagosklonny, 2013).

These frameworks have transformed the field. They have generated vocabulary, hypotheses, experimental models, and therapeutic programs. Yet as aging biology becomes increasingly connected to drug discovery, a different kind of theoretical object is required. It is no longer sufficient to ask only which processes change with age. A therapeutic science of aging must answer a more operational question:

This question is not answered by a list of Hallmarks, a catalogue of damage classes, or a metaphor of information loss. Those frameworks identify relevant biology, but they do not by themselves specify the dynamics of the aged system, the state-dependent response to intervention, the safety-constrained set of possible actions, or the objective function that defines therapeutic success. They tell us what may matter. They do not fully specify what to do next.

The distinction is analogous to a historical transition in physics. Thermodynamics began as an empirical science of heat, work, pressure, and engines. It classified regularities and supplied phenomenological laws. Yet a deeper dynamical account required statistical mechanics: a description of how macroscopic phenomena arise from the behavior of microscopic states. In a similar sense, the Hallmarks of Aging represent a powerful phenomenological ontology. They organize observations about aging systems. But drug discovery requires a dynamical and interventional layer: equations of motion for biological states, explicit control inputs, response operators, constraints, objective functions, and predictions that can fail.

The purpose of this paper is to propose such a layer. We do not argue that existing theories are wrong. On the contrary, the framework developed here depends on their insights. Evolutionary theory explains why aging can exist; damage theories and SENS identify lesion classes; Hallmarks organize major mechanisms; information theories identify potentially reversible state variables; geroscience provides translational context; hyperfunction theory highlights growth and nutrient signaling as control-relevant axes. Our contribution is narrower but, we argue, operationally important: we formulate aging as a problem of safe controllability and define biological age as the minimum safe control cost required to maintain or restore function.

Control theory is the natural mathematical language for this problem. In engineering, control theory addresses how dynamical systems can be guided from one state to another under constraints, uncertainty, costs, and limited actuators (Kalman, 1960; Bellman, 1957; Kirk, 1970; Sontag, 1998). The central objects of control theory — state variables, dynamics, inputs, response operators, reachable sets, value functions, constraints, and optimal policies — map directly onto the needs of interventional biology. A biological system has internal state variables. It evolves over time under endogenous dynamics and stochastic perturbations. Drugs, cell therapies, genetic perturbations, nutritional interventions, exercise, environmental changes, and surgical procedures act as inputs. Their effects are state-dependent: the same intervention can be beneficial, neutral, or harmful depending on the biological context. Therapeutic success is not movement toward an abstract youthful ideal but restoration or maintenance of function under acceptable risk.

Aging drug discovery is therefore not merely a target-identification problem. It is a control problem under uncertainty. The relevant question is not simply whether a gene is associated with age, whether a pathway belongs to a Hallmark, or whether an intervention reverses a biomarker. The relevant question is whether manipulating that target changes the trajectory of an aged system in a direction that reduces the minimum cost of functional restoration. A drug is not merely "anti-inflammatory," "senolytic," "rapalog," or "epigenetic." In the formalism proposed here, a drug is a vector field on biological state space: a state-dependent transformation operator that pushes the system along particular directions. A target is valuable if its modulation reduces the optimal control cost of returning the system to a functional viability set. A combination is valuable if its component vector fields expand the reachable safe set more than either intervention alone. A sequence matters when the vector fields do not commute.

This view also clarifies the limitations of biological age measures. Current biological clocks, especially epigenetic clocks, have provided remarkable predictors of chronological age, disease risk, and mortality (Horvath, 2013; Hannum et al., 2013; Levine et al., 2018; Lu et al., 2019). However, a clock is not necessarily an intervention law. A biomarker can predict age without indicating how to restore function. A methylation signature may move in a youthful direction without improving tissue architecture, resilience, or survival. Conversely, an intervention may improve function without strongly reversing a canonical clock. A control-theoretic definition of biological age shifts emphasis from correlation to actionability. A system is biologically older if it requires more intervention energy, greater risk, more complex combinations, or longer time to restore function — or if no safe intervention path exists.

Why is such a framework plausible now? Three developments make it timely. First, multi-omics and single-cell technologies increasingly permit high-dimensional measurement of biological state. Transcriptomic, proteomic, metabolomic, epigenomic, imaging, histological, and clinical data can be integrated into latent representations of tissue and organismal state. Second, perturbational datasets — including CRISPR screens, LINCS/CMap profiles, drug-response atlases, organoid perturbations, and animal intervention studies — provide empirical information about how biological systems respond to inputs. Third, artificial intelligence and machine learning can estimate complex state representations, infer dynamics, rank targets, model response heterogeneity, and integrate proprietary and public datasets across disease areas. These technologies do not eliminate the need for theory. Rather, they make a mathematically explicit theory of interventional aging possible.

Four Central Commitments

First, aging is described in state space. Let x(t) ∈ ℝⁿ represent the latent biological state of a tissue, organ system, or organism. This state may include epigenetic configuration, gene expression, protein abundance, metabolic flux, cellular composition, extracellular matrix architecture, immune tone, organ reserve, and functional performance. Observed biomarkers y(t) are incomplete and noisy functions of this latent state:

Second, health is defined not as youth but as functional viability. We define a functional viability set,

where F_i(x) are functions such as cardiac reserve, renal filtration, immune competence, muscle strength, cognitive performance, wound healing, and metabolic flexibility. The aim of intervention is not to make a 70-year-old identical to a 20-year-old at every molecular coordinate. The aim is to maintain or restore adequate function safely.

Third, interventions are controls. The uncontrolled aging system evolves according to

where a is chronological age, e environment, g genotype, f endogenous drift, and Σ(x_t) dW_t stochastic perturbation. Interventions add controlled vector fields:

Here u_j(t) is the dose or intensity of intervention j, and g_j(x) describes the direction and magnitude of its state-dependent effect. This formulation immediately explains why the same drug can have different effects in young and old systems, in inflamed and non-inflamed tissues, in fibrotic and non-fibrotic organs, or in resilient and frail organisms.

Fourth, biological age is defined by a constrained value function. We define the control biological age of a state as a monotone function of the minimum safe cost required to maintain or restore function:

where

Here ℓ(x_t) penalizes functional loss or distance from the viability set, c(u_t) penalizes intervention burden, r(x_t, u_t) penalizes toxicity and safety risk, and Φ(x_T) is a terminal penalty. Young, resilient systems have low restoration cost. Aged systems have higher cost. Severely damaged systems may lie outside the safe controllable region, such that no admissible intervention can restore function.

This formulation is not only conceptual. It generates concrete predictions. It predicts that some interventions that improve Hallmark-associated biomarkers will fail if their vector fields do not project onto deficient controllable modes. It predicts that high-value targets may lie outside canonical Hallmark annotations if they control state transitions or recovery rates. It predicts that intervention order matters when drug-induced vector fields do not commute. It predicts that perturbation-response measures of resilience can detect loss of controllability before static biomarkers become abnormal. It predicts that partial reprogramming has a calculable ceiling in structurally aged tissues, where fibrosis, calcification, clonal expansion, or tissue architecture constrain the reachable set. It predicts state-dependent sign reversals, such that suppressing inflammation, mTOR, senescence, or proliferation can be beneficial in one control regime and harmful in another.

Finally, the framework is designed to be falsifiable. A useful drug-discovery theory should outperform simpler alternatives. We therefore propose validation against a corpus of 30 AI-discovered preclinical candidate compounds, scored by Hallmark annotation, age association, disease association, network centrality, druggability, toxicity risk, and estimated control-value reduction. The central empirical hypothesis is that control-value scores predict preclinical advancement, functional efficacy, and toxicity-adjusted therapeutic index better than Hallmark membership alone. This analysis can be performed retrospectively within AI-driven drug discovery pipelines and prospectively in independent experimental systems.

The sections that follow develop the background, formalism, drug-as-vector-field concept, a fully specified five-dimensional worked example of aged murine liver, connections to network controllability and systems pharmacology, identifiability and robust-control analysis, and a catalogue of twenty novel predictions.

2.Background: The Landscape of Aging Theory

Aging is not a single observation but a hierarchy of phenomena: molecular damage, cellular dysregulation, tissue remodeling, organ decline, increased disease susceptibility, reduced resilience, and rising mortality risk. No single existing theory explains all levels with equal precision. Instead, the field has developed complementary frameworks, each addressing different questions. Evolutionary theories ask why aging exists. Damage theories ask what accumulates. Hallmarks classify recurring mechanisms. SENS proposes repair targets. Information theories emphasize epigenetic and regulatory reversibility. Geroscience translates aging biology into disease prevention. Hyperfunction theory identifies persistent growth programs as causal drivers. Each contributes essential insight. Yet none alone provides a full control law for intervention.

2.1 Evolutionary theories: why aging can exist

Modern aging theory begins with the recognition that natural selection weakens with age. Medawar proposed that deleterious late-acting mutations can accumulate because selection is less effective after reproduction (Medawar, 1952). Williams extended this idea through antagonistic pleiotropy: alleles beneficial early in life may be favored even if harmful later (Williams, 1957). Hamilton formalized the declining force of natural selection with age, providing a mathematical foundation for the evolution of senescence (Hamilton, 1966). Kirkwood's disposable soma theory further argued that organisms allocate limited resources between reproduction and somatic maintenance; aging emerges because evolution optimizes fitness, not indefinite repair (Kirkwood, 1977; Kirkwood and Holliday, 1979).

These theories remain fundamental. They explain why late-life decline is compatible with natural selection. They also predict trade-offs between growth, reproduction, maintenance, and repair. However, evolutionary theories do not specify which intervention should be used in a given aged tissue. They do not define the state variables of an individual organism, the dose-response operator of a drug, or the safety-constrained trajectory back to function. Evolutionary logic can tell us why aging mechanisms were not eliminated by selection; it does not by itself identify the optimal therapeutic control policy for a 72-year-old patient with fibrotic lung remodeling, immunosenescence, and reduced renal reserve.

2.2 Damage theories: what accumulates

Damage theories emphasize the progressive accumulation of molecular and cellular lesions. Harman's free radical theory proposed that reactive oxygen species contribute to aging by damaging macromolecules (Harman, 1956). Later mitochondrial theories focused on mitochondrial DNA mutations, respiratory chain dysfunction, and ROS production as contributors to age-related decline (Harman, 1972; Wallace, 2005). Other damage theories emphasize DNA lesions, protein aggregates, lipofuscin, advanced glycation end products, extracellular matrix crosslinks, and somatic mutations.

Damage theories identify plausible substrates of functional decline. They also motivate repair, replacement, or removal strategies. Nevertheless, damage burden alone does not specify intervention priority. If an aged tissue contains DNA damage, senescent cells, mitochondrial dysfunction, fibrosis, inflammatory signaling, and stem-cell exhaustion, which should be addressed first? Which lesions are causal bottlenecks rather than correlated byproducts? Which interventions restore function rather than merely normalize biomarkers? Which damage classes are still controllable under safety constraints? A damage inventory is necessary but insufficient for optimal intervention.

2.3 SENS: repair catalogues and the need for dynamics

The Strategies for Engineered Negligible Senescence program advanced an explicitly interventionist view. Rather than attempting to slow all causes of damage, SENS proposed periodically repairing or removing specific categories of accumulated damage, including cell loss, death-resistant cells, extracellular aggregates, intracellular aggregates, mitochondrial mutations, extracellular crosslinks, and cancer-promoting nuclear mutations (de Grey et al., 2002; de Grey, 2007). This was a major conceptual shift. It framed aging as an engineering problem and emphasized that therapies could target damage even without complete knowledge of upstream causes.

The control-theoretic framework developed here is sympathetic to that engineering orientation. However, SENS is primarily a catalogue of damage-repair categories rather than a quantitative dynamics of tissue state under intervention. It does not generally specify equations governing how different damage classes interact, how repair of one lesion changes the controllability of another, how dose and timing should be optimized, or how safety constraints reshape the feasible intervention set. For example, clearing senescent cells may improve tissue function in SASP-dominated environments but impair wound repair if transient senescence is serving a regenerative role (Demaria et al., 2014; Ritschka et al., 2017). Repair logic must therefore be embedded in state-dependent dynamics.

2.4 The Hallmarks of Aging: a dominant ontology, not a control model

The Hallmarks of Aging framework is arguably the most influential organizing schema in contemporary aging biology (López-Otín et al., 2013; López-Otín et al., 2023). It integrated diverse observations into a coherent set of mechanisms, each satisfying criteria of age association, experimental aggravation, and therapeutic amelioration. The Hallmarks framework has been highly productive: it has guided experiments, reviews, grant programs, target discovery, and geroscience translation.

Yet the Hallmarks are best understood as a biological ontology, not a complete dynamical theory of intervention. They define categories of change. They do not define a state vector x(t), equations of motion f(x), intervention vector fields g_j(x), dose-response functions, safety constraints, objective functions, reachable sets, or optimal policies. They do not tell us whether, in a given aged biological state, mTOR inhibition should precede senolysis, whether extracellular matrix remodeling should precede reprogramming, whether immune suppression will reduce pathology or impair host defense, or whether a target outside the Hallmark vocabulary may have greater control leverage than a canonical Hallmark node.

This is not a criticism of the Hallmarks as an ontology. Lists are indispensable in fields with complex mechanisms. But a list of mechanisms is not the same as a theory of controlled dynamics. A Hallmark annotation can identify that a target is related to cellular senescence, mitochondrial dysfunction, or inflammation. It cannot by itself quantify whether modulating that target reduces the minimum safe cost of restoring organ function. As aging biology enters therapeutic development, this distinction becomes increasingly important.

2.5 Information theories and reprogramming: reversibility and its limits

Information-loss theories emphasize that aging involves disruption of regulatory information, cellular identity, chromatin organization, and epigenetic state (Sinclair and LaPlante, 2019; Yang et al., 2023). Experimental partial reprogramming has demonstrated that some age-associated cellular features can be reversed, at least in specific contexts (Ocampo et al., 2016; Lu et al., 2020; Gill et al., 2022). These findings are among the most important developments in modern aging biology because they show that aging is not purely cumulative irreversible damage; some components are dynamic and resettable.

However, reprogramming also illustrates the need for control theory. Epigenetic rejuvenation is not equivalent to functional restoration in all tissues. A cell may acquire a younger molecular signature while remaining embedded in a fibrotic extracellular matrix, deprived of vascular supply, constrained by tissue architecture, or surrounded by chronic inflammation. Conversely, aggressive reprogramming may increase cancer risk, erase cell identity, or cause loss of function. Thus, reprogramming is a control input with a narrow safety envelope, state-dependent efficacy, and structural constraints on its reachable set.

A control-theoretic framework generalizes information theories by treating epigenetic state as one component of x(t), not the entire state. It can ask: when does epigenetic control reduce restoration cost? When is it insufficient because structural variables dominate? When should it be preceded by senolysis, antifibrotic remodeling, metabolic stabilization, or immune modulation? What is the optimal pulse duration? What safety constraints define the admissible set? These are not questions of metaphor but of controlled dynamics.

2.6 Geroscience: translational ambition without formal optimization

Geroscience proposes that targeting fundamental mechanisms of aging can prevent or delay multiple chronic diseases (Kennedy et al., 2014; Sierra and Kohanski, 2017; Barzilai et al., 2018). This field has driven clinical interest in rapalogs, metformin, senolytics, NAD metabolism, anti-inflammatory interventions, immune rejuvenation, and other approaches. It has also reframed aging as a modifiable risk factor rather than an inevitable background process.

The translational strength of geroscience is precisely why optimization becomes necessary. Clinical intervention requires choices: which population, which endpoint, which mechanism, what dose, what duration, what combination, and what safety trade-off? Chronological age is an imperfect enrollment criterion. Biomarker age may not indicate intervention response. A geroscience trial must decide whether to target relatively healthy older adults, pre-frail individuals, patients with age-related disease, or those with specific molecular profiles. These are control questions. The success of geroscience will depend on moving from broad mechanism classes to quantitative response prediction.

2.7 Hyperfunction theory: mechanistic specificity and pathway control

Hyperfunction theory, especially as developed around mTOR signaling, argues that aging is driven not only by passive damage accumulation but by persistent activity of growth-promoting pathways after development (Blagosklonny, 2006; Blagosklonny, 2013). This framework explains why growth, nutrient sensing, and anabolic signaling can become pathological in later life. It also aligns with evidence that rapamycin extends lifespan in multiple model organisms and can improve specific age-related phenotypes (Harrison et al., 2009; Miller et al., 2011; Kennedy and Lamming, 2016).

Hyperfunction theory has strong mechanistic and therapeutic content. Yet aging is unlikely to be reducible to one pathway. mTOR inhibition may improve some states but worsen others, particularly where regeneration, immune competence, or wound healing are limiting. The control-theoretic framework incorporates hyperfunction as a specific vector field or family of vector fields in state space. It asks when mTOR inhibition reduces the value function and when it increases it. In other words, rapamycin is not assigned a fixed sign. Its sign depends on the state-dependent projection of its vector field onto the gradient of restoration cost.

2.8 Reliability, resilience, and critical transitions

Reliability theory and resilience frameworks provide additional foundations. Gavrilov and Gavrilova modeled aging as progressive failure of redundant system components, producing mortality dynamics that can resemble Gompertz-like behavior (Gavrilov and Gavrilova, 1991; Gavrilov and Gavrilova, 2001). Resilience theory, developed in ecology and complex systems, shows that systems approaching critical transitions often exhibit slower recovery, increased variance, and reduced stability before overt collapse (Scheffer et al., 2009; Scheffer et al., 2012). These ideas map closely onto aging, where frailty and disease often emerge after a period of declining reserve.

Control theory extends resilience concepts by incorporating interventions. It is not enough to measure that recovery is slowing. We must ask what inputs can restore recovery, how much they cost, and whether safe paths exist. The controllability Gramian, reachable sets, and value functions provide mathematical tools for quantifying this loss of resilience in interventional terms. Frailty-index approaches (Mitnitski et al., 2002; Rockwood and Mitnitski, 2007) and universal biological-age markers of stress and frailty (Pyrkov and Fedichev, 2019) provide empirical entry points for operational biological age.

2.9 What existing frameworks lack in common

Despite their differences, most aging frameworks lack several components required for drug discovery:

1. State variables. What is the mathematical state of the biological system?
2. Dynamics. How does the state evolve without intervention?
3. Control inputs. What can be manipulated?
4. Response operators. How does each intervention affect the state, and how does that effect depend on context?
5. Constraints. Which interventions are unsafe, infeasible, or toxic?
6. Objective function. What counts as improvement?
7. Optimality conditions. Which intervention or sequence minimizes cost while restoring function?
8. Falsifiable predictions. What quantitative outcomes would refute the framework?

The absence of these elements does not make previous theories wrong. It indicates that they were built to answer different questions. Drug discovery, however, requires a theory of safe control. The next section formalizes this theory.

3.The Control-Theoretic Framework

3.1 Biological state space

We begin by representing a biological system as a latent state vector,

The state may describe a cell population, tissue, organ, physiological subsystem, or whole organism. It is latent because no experimental platform measures all biologically relevant variables. Instead, observed data are noisy projections:

where y(t) may include transcriptomic, proteomic, metabolomic, epigenetic, imaging, histopathological, clinical, and functional measurements. The components of x(t) need not correspond one-to-one to measured biomarkers. In practice, x(t) may be inferred as a low-dimensional or structured representation learned from multi-omics, clinical, and perturbational data. Its dimensions may include:

• epigenetic configuration and chromatin accessibility;
• transcriptional and proteomic regulatory state;
• metabolic flux and nutrient-sensing state;
• mitochondrial function and redox balance;
• proteostasis capacity;
• DNA damage and repair competence;
• senescent-cell burden and SASP intensity;
• immune activation, exhaustion, and surveillance;
• stem-cell reserve and differentiation capacity;
• extracellular matrix organization and stiffness;
• fibrosis, calcification, and tissue architecture;
• vascular supply and endothelial function;
• organ reserve and physiological performance;
• frailty, mobility, cognition, and systemic function.

The state can be tissue-specific or systemic. A kidney state vector may emphasize nephron number, filtration barrier integrity, tubular stress, inflammation, fibrosis, and vascular function. A lung state vector may emphasize epithelial injury, fibroblast activation, extracellular matrix remodeling, immune infiltration, gas exchange, and compliance. A systemic state vector may integrate immune tone, metabolic state, endocrine signaling, vascular health, body composition, and functional reserve.

This distinction is important because aging is neither purely cell-autonomous nor purely systemic. Tissue-specific states interact through circulating factors, immune trafficking, neuroendocrine regulation, microbiome-derived metabolites, and behavior. A practical framework may therefore use hierarchical state representations:

The correct level of description depends on the intervention and endpoint. A senolytic in fibrotic lung disease may require tissue-specific state estimation, whereas an immune-modulatory intervention may require systemic and tissue compartments.

3.2 The functional viability set

Aging research often describes intervention goals as "rejuvenation," "youthfulness," or reversal of biological age. These terms can be useful heuristics, but they are imprecise therapeutic objectives. The goal of medicine is not to make every molecular feature identical to youth. It is to preserve or restore function safely.

We therefore define a functional viability set,

where each F_i(x) is a functional measure and θ_i is an acceptable threshold. Examples include cardiac reserve under stress, renal filtration and tubular concentrating capacity, immune competence against infection and malignancy, wound healing, muscle force and mobility, cognitive performance, pulmonary compliance and gas exchange, hematopoietic output and immune repertoire diversity, metabolic flexibility, and vascular reactivity.

The thresholds θ_i need not be those of a young adult. They may be age-appropriate, disease-specific, or patient-specific. For example, restoring an older patient with chronic kidney disease to safe filtration and electrolyte handling may be a valid success even if nephron number is not restored to young-adult levels. Similarly, improving immune competence without inducing autoimmunity may be preferable to maximizing immune activation.

The viability set also avoids an error common in rejuvenation discourse: treating youth as an attractor. There is no guarantee that youth is a natural attractor of biological dynamics. Developmental trajectories are not simply reversible aging trajectories. A control objective based on function is therefore more biologically and clinically appropriate.

A related concept is the viability kernel: the set of states from which there exists at least one admissible control policy capable of maintaining the system within 𝒱 over a specified horizon. Formally, the viability kernel 𝒦 is

States outside the viability kernel may still be improved, but they cannot be maintained within functional bounds under available safe interventions. This provides a formal language for frailty, irreversibility, and late-stage disease.

3.3 Aging dynamics without intervention

In the absence of intervention, the biological state evolves according to stochastic dynamics:

Here f is the endogenous drift; a denotes chronological age; e environment; g genotype; Σ(x_t) the state-dependent noise matrix; and W_t a Wiener process capturing stochastic fluctuations. The inclusion of chronological age does not imply that age is an independent causal force. Rather, a can index time-dependent exposures, developmental history, cumulative damage, and changing regulatory regimes.

Aging corresponds to several changes in these dynamics.

First, the drift term increasingly moves the system away from the viability set. This drift can arise from damage accumulation, chronic inflammation, dysregulated nutrient sensing, clonal expansion, extracellular matrix remodeling, stem-cell exhaustion, and other mechanisms. In young systems, homeostatic feedback often returns perturbed variables toward functional ranges. In aged systems, feedback becomes slower, weaker, or maladaptive.

Second, stochastic instability increases. Biological aging is associated with increased transcriptional noise, epigenetic drift, proteostatic stress, mitochondrial heterogeneity, immune repertoire contraction, clonal mosaicism, and physiological variability (Bahar et al., 2006; Enge et al., 2017; Martinez-Jimenez et al., 2017). In control terms, Σ(x) increases or becomes more destabilizing in certain regions of state space.

Third, recovery after perturbation slows. A resilient system returns rapidly to baseline after infection, injury, metabolic challenge, or stress. An aged system may recover incompletely or transition to a new pathological attractor. Slower recovery is a known early warning signal in complex systems approaching critical transitions (Scheffer et al., 2009). In aging, it may be a more informative indicator than static biomarkers.

Fourth, the dimensionality of safe control decreases. A young tissue may respond to many interventions because its repair pathways, stem-cell reserves, vascular supply, and immune functions remain intact. An old tissue may have fewer safe directions of movement. Some interventions that were beneficial earlier become ineffective or toxic because compensatory systems have failed.

Finally, the cost of restoration increases. The system may still be movable, but only with higher drug burden, greater toxicity, longer treatment, or combinations. At advanced stages, no safe path may exist.

Classical mortality patterns, including Gompertz-like exponential increases in mortality with age, may emerge from these dynamics if the probability of crossing viability boundaries increases as drift, noise, and loss of redundancy accumulate (Gompertz, 1825; Gavrilov and Gavrilova, 1991; Gavrilov and Gavrilova, 2001). The present framework does not require deriving the Gompertz law from first principles, but it is compatible with reliability and resilience accounts of rising hazard.

3.4 Intervention as control

Interventions enter the dynamics as control inputs:

or in matrix notation,

Here u_j(t) denotes the intensity, dose, schedule, or exposure of intervention j, and g_j(x) is the corresponding intervention vector field. The matrix G(x) collects all available intervention vector fields.

This equation is the core translation from aging biology to drug discovery. A drug is not merely a label or pathway inhibitor. It is a transformation of biological state. For example:

• a rapalog may reduce mTOR-dependent anabolic signaling, alter autophagy, modulate immune function, and affect regeneration;
• a senolytic may reduce senescent-cell burden and SASP signaling, but may also affect repair-associated senescent cells;
• an antifibrotic may remodel extracellular matrix dynamics and fibroblast activation;
• a reprogramming pulse may alter epigenetic age, cellular identity, proliferation risk, and differentiation potential;
• an immune modulator may reduce chronic inflammation while impairing host defense;
• a metabolic intervention may shift nutrient sensing, mitochondrial flux, and systemic energy allocation.

The vector field g_j(x) is state-dependent. This is not a technical detail; it is biologically central. The same drug can push different states in different directions. Rapamycin in a metabolically overactive, inflammatory, pre-frail state may reduce pathology. Rapamycin in a frail state requiring wound repair may impair recovery. Senolytics in a tissue with high chronic senescent burden may restore function. Senolytics during acute wound healing may remove cells contributing to repair. Partial reprogramming in a structurally intact tissue may improve function. Partial reprogramming in a fibrotic or neoplastic-prone tissue may fail or create risk.

The sign of an intervention is therefore not intrinsic. It depends on the value function. Locally, intervention j is beneficial when

meaning the vector field points in a direction that decreases restoration cost. It is harmful when

This formalizes context dependence.

3.5 The core definition: biological age as control cost

We now define the optimal control value function:

subject to

The terms have explicit biological interpretation:

• ℓ(x_t) penalizes functional loss, distance from 𝒱, or risk of leaving 𝒱;
• c(u_t) penalizes intervention burden, including dose, duration, invasiveness, cost, and patient burden;
• r(x_t, u_t) penalizes toxicity and safety risk;
• Φ(x_T) penalizes terminal dysfunction or failure to reach the target region;
• 𝒰_safe is the set of admissible interventions satisfying safety, feasibility, pharmacologic, and ethical constraints.

A simple choice for ℓ(x) is distance from the viability set, ℓ(x) = d(x, 𝒱), or a weighted sum of functional deficits, ℓ(x) = Σ_i w_i max(0, θ_i − F_i(x)).

The value function V(x₀, T) is the minimum expected cost of maintaining or restoring function over horizon T. We define control biological age as

where φ maps restoration cost to an age-equivalent or clinically interpretable scale.

This definition differs from chronological age, biomarker age, and Hallmark burden. Chronological age measures time lived. Biomarker age predicts age or outcomes from molecular features. Hallmark burden counts or scores age-associated processes. Control biological age measures how difficult it is to restore or maintain function safely.

A young biological system has low V. If perturbed, it can return to function through endogenous repair or modest intervention. A middle-aged system has higher V, requiring stronger or more targeted intervention. An old but robust system may have moderate V if function is maintainable. A frail system has high V, reflecting narrow safety margins and limited reserve. A terminally damaged system may have effectively infinite V if no admissible control path reaches 𝒱.

3.6 Local linearization and practical computation

Although biological systems are nonlinear and stochastic, local approximations can be useful. Around a state or trajectory, one may approximate dynamics as

where A(a) is the age-dependent endogenous dynamics matrix and B(a) is the intervention susceptibility matrix. The controllability Gramian over time horizon T is

For linear systems, the minimum energy required to move from x₀ to x_T is

This approximation yields testable predictions. With age, disease, or structural damage, the smallest eigenvalues of W should decline, λ_min(W(T, a)) ↓, and the minimum restoration energy should increase, E*(x₀, x_T, a) ↑. These quantities can be estimated from perturbation-response data, longitudinal omics, organoid models, animal studies, and clinical time series. They provide a bridge between abstract control theory and empirical aging biology.

4.Drugs as Vector Fields

4.1 The drug-as-vector-field concept

Drug discovery often describes compounds by their primary target: "rapamycin inhibits mTOR," "dasatinib inhibits tyrosine kinases," "metformin activates AMPK," "a senolytic kills senescent cells," or "a reprogramming factor resets epigenetic state." Such descriptions are useful but incomplete. They compress a high-dimensional intervention into a scalar mechanism.

In the control-theoretic framework, a drug induces a vector field g_j(x) on biological state space. Its effect depends on the current state, dose, duration, tissue context, pharmacokinetics, pharmacodynamics, and interactions with endogenous feedback. Under intervention u_j(t), the trajectory changes according to

This representation provides a natural language for responder heterogeneity. Two individuals may have the same chronological age but different x. The same drug may reduce V in one and increase V in another. Similarly, the same person may respond differently at different times, depending on inflammatory state, infection, nutritional status, tissue injury, microbiome, or comorbid disease.

The drug-as-vector-field concept also clarifies dose-response. A low dose may move the system gently along a beneficial direction. A high dose may push the state into toxicity or activate compensatory feedback. A pulsed intervention may allow recovery between state transitions. A continuous intervention may suppress necessary repair. Optimal dosing is therefore not simply maximizing target engagement. It is selecting u(t) to minimize the value function under constraints.

This view is especially important for aging interventions because many target homeostatic pathways rather than pathogen-specific mechanisms. mTOR, AMPK, insulin/IGF signaling, inflammation, autophagy, senescence, and epigenetic regulation all serve beneficial roles in some contexts. Their therapeutic value depends on state-dependent control.

4.2 Non-commutativity: why order matters

In many biomedical frameworks, combinations are treated as additive or synergistic sets of mechanisms. If one intervention targets senescence and another targets epigenetic state, the combination may be expected to address two Hallmarks. But control theory predicts that sequence can matter independently of the components. Intervention A followed by intervention B may not equal intervention B followed by intervention A.

Mathematically, two vector fields g_A and g_B commute only if their Lie bracket is zero:

When [g_A, g_B] ≠ 0, the order of interventions changes the trajectory. Thus, A → B ≠ B → A.

This is not an abstract mathematical curiosity. Biological interventions frequently alter the state variables that determine response to subsequent interventions. A senolytic may reduce inflammatory signaling and remove cells that constrain tissue remodeling, thereby changing the response to reprogramming. Conversely, reprogramming before senolysis may act on a tissue still dominated by SASP, fibrosis, or immune dysfunction, increasing risk or reducing efficacy. Antifibrotic remodeling may improve tissue architecture and allow regenerative therapy to act. Regenerative stimulation before matrix remodeling may fail because cells cannot organize into functional tissue. Metabolic stabilization may reduce stress and improve the safety of epigenetic modulation. Epigenetic modulation in an unstable metabolic state may increase dysfunction.

These sequence effects are not generally predicted by Hallmark annotation alone. A Hallmark framework can suggest combining senolysis and reprogramming; it does not define when senolysis should precede reprogramming or vice versa. The control framework predicts order from the non-commutativity of intervention vector fields and the value function landscape. Section 6 makes this concrete for a five-dimensional model of aged murine liver, in which the Lie bracket of senolytic and reprogramming vector fields is explicitly computed and shown to be non-zero.

This yields a direct experimental test. For a pair of interventions A and B, estimate g_A, g_B, and their Lie bracket in a relevant aged tissue model. Predict whether A → B, B → A, simultaneous treatment, or monotherapy best reduces V. Then test predefined functional endpoints, safety markers, and durability. If strong predicted non-commutativity repeatedly fails to correspond to sequence-dependent outcomes, the model is wrong.

4.3 Controllability and its loss

Controllability asks whether a system can be moved from its current state to a target set using available controls. In aging biology, the relevant question is whether the current state can be moved into or maintained within the functional viability set 𝒱 using admissible interventions.

Aging can reduce controllability in several ways.

First, intervention susceptibility may decline. The matrix B(a) or vector fields g_j(x) may weaken with age. For example, stem-cell exhaustion can reduce response to regenerative signals; vascular rarefaction can reduce drug delivery and repair; immune exhaustion can reduce response to vaccination or immunotherapy.

Second, the target set may become harder to reach because structural variables change. Fibrosis, extracellular matrix crosslinking, calcification, neuronal loss, nephron loss, sarcopenia, and tissue architectural collapse can create barriers that cannot be reversed by transcriptional modulation alone.

Third, safety constraints tighten. An intervention dose tolerated by a robust adult may be unsafe in frailty, renal impairment, immunosuppression, or cancer predisposition. Thus the admissible control set 𝒰_safe shrinks with age and comorbidity.

Fourth, stochasticity increases. Greater noise makes trajectories less predictable and raises the risk of crossing unsafe boundaries.

These processes create the possibility of irreversible transitions. A tissue may cross a boundary beyond which no safe intervention can restore function. Let 𝒞 denote the controllable set:

The boundary ∂𝒞 formalizes the "point of no return." This boundary is not necessarily fixed. New therapies can expand 𝒞. Combination therapies can expand 𝒞 if their vector fields open safe paths unavailable to monotherapies. But for any given therapeutic toolkit, some states may be unreachable.

It is also useful to distinguish the full controllable set 𝒞 from the viability-constrained reachable set:

ℛ_safe excludes trajectories that transit through unsafe regions even if they arrive at viable endpoints. This distinction matters for reprogramming, which may move epigenetic coordinates toward youth while traversing dedifferentiated or tumorigenic intermediate states.

This concept is clinically important. Many interventions fail because they are applied after the system has crossed a controllability threshold. For example, epigenetic reprogramming may improve molecular markers in a tissue where architecture is too damaged to restore function. Anti-inflammatory therapy may fail in end-stage fibrosis because inflammation is no longer the limiting mode. Senolytics may fail when tissue loss, rather than senescent burden, dominates the state. A control framework distinguishes mechanisms that remain actionable from mechanisms that are merely present.

4.4 Safety constraints

Safety is not an external consideration added after efficacy. In aging interventions, safety is part of the control problem. The admissible set 𝒰_safe includes only interventions, doses, schedules, and combinations that do not create unacceptable harm. Examples include cancer risk from reprogramming or proliferative stimulation; immune suppression from rapalogs, corticosteroids, or anti-inflammatory agents; impaired wound healing from excessive senolysis or mTOR inhibition; bleeding, thrombocytopenia, or off-target toxicity from senolytics; metabolic decompensation from nutrient-sensing interventions; arrhythmia, renal injury, or hepatic toxicity from systemic drugs; clonal expansion or immune escape from interventions affecting cell turnover.

The safety-efficacy trade-off can be represented as a constraint surface. A high-dose intervention may have a large vector field toward the viability set but cross a toxicity boundary. A lower-dose combination may achieve a similar movement with less risk because each component contributes along complementary directions. Thus, combinations can be safer than aggressive monotherapy, not merely more effective.

Formally, the optimal policy solves

subject to u(t) ∈ 𝒰_safe and x_t ∉ 𝒳_unsafe. Here 𝒳_unsafe may include oncogenic states, immune collapse, severe inflammation, organ failure, or unacceptable functional loss. The safest intervention is not always the weakest. It is the one that moves the system efficiently while avoiding unsafe regions of state space.

4.5 Relationship to Systems Pharmacology

The idea that drugs can be represented as state perturbations has substantial prior art in pharmacology, systems biology, and computational drug discovery. The present framework does not claim priority for that general idea. Rather, it applies and extends it to aging by embedding drug-induced state perturbations inside an aging-specific dynamical system with explicit safety constraints, age-dependent controllability, and a value-function definition of biological age.

Classical pharmacokinetic/pharmacodynamic modeling already used state-space language. In the work of Sheiner, Holford, and colleagues, drug concentration and effect compartments were represented by dynamical systems in which dosing regimens changed physiological outputs through parameterized pharmacodynamic relationships. These models established the modern quantitative basis for dose-response prediction, therapeutic windows, and individualized dosing. Systems pharmacology later generalized this idea from single drug-effect compartments to interacting biochemical and cellular networks. Sorger, Iyengar, Bhalla, and others emphasized that drug action is best understood as perturbation of a biological network rather than as modulation of an isolated target.

The Connectivity Map provided an especially important empirical realization of this principle. Lamb et al. (2006) showed that small molecules could be represented by gene-expression signatures and that disease and drug signatures could be compared in transcriptomic space. Subramanian et al. (2017) extended this approach through the L1000 platform, scaling perturbational transcriptomics to large numbers of compounds, genes, and cellular contexts. In the language of the present paper, CMap-like methods estimate empirical perturbation directions in high-dimensional molecular state space. A compound is not merely a ligand for a target; it is an operator that moves a cell state along a characteristic transcriptional direction.

Our framework is therefore continuous with systems pharmacology, not a replacement for it. The difference lies in the object being controlled and in the mathematical criterion used to evaluate control. CMap asks whether a compound reverses a disease-associated signature or induces a desired molecular profile. PK/PD asks what dose produces what effect over what time course. Network pharmacology asks how target modulation propagates through interacting pathways. The present framework asks a related but distinct question: given an aged biological state x₀, what sequence of interventions minimizes the expected cost of moving the system toward a functional youthful state while remaining inside a viability set?

There are four specific extensions.

First, we include aging-specific dynamics. Aging is not a static disease signature but a trajectory. The same perturbation may have different effects depending on senescent burden, regenerative capacity, epigenetic integrity, fibrosis, inflammation, and accumulated damage. Therefore a drug is represented not simply as a vector in molecular space but as a controlled vector field g_j(x) whose effect depends on the current state.

Second, we impose safety and viability constraints. Many perturbations can move molecular signatures toward youth in vitro while being unacceptable in vivo. Partial reprogramming illustrates this point: it can reverse epigenetic age markers, but excessive or poorly timed reprogramming can cause loss of cellular identity, teratoma formation, or oncogenic risk. The relevant object is therefore not an unconstrained perturbation vector but an admissible controlled vector field acting within 𝒱, the biologically viable region of state space.

Third, we explicitly represent controllability loss over time. Aging changes not only the state but also the ease with which the state can be modified. Fibrosis, clonal hematopoiesis, stem-cell exhaustion, immune remodeling, and accumulated damage can reduce the effectiveness or safety of interventions. A drug vector field estimated in young cells may not extrapolate to old tissue. Thus, the framework requires state-conditioned pharmacodynamics rather than age-invariant signatures.

Fourth, we define biological age through a value function. Epigenetic clocks estimate position along an aging-associated molecular axis. Our proposed BA_control instead measures the minimum safe cost of functional restoration. This is not offered as a currently mature clinical biomarker, but as an operational research definition: a state is biologically older if it is more costly, riskier, or less feasible to return it to a youthful functional region.

4.6 Relationship to Network Controllability

The present framework is closely related to, but distinct from, the network controllability program initiated by Liu, Slotine, and Barabási (2011), who showed that the structural controllability of a directed complex network can be characterized from network topology and that the minimum number of driver nodes is determined by maximum matching structure rather than by node degree alone. In that formulation, a linearized network system is written as

where x ∈ ℝⁿ is the network state, A encodes directed interactions among nodes, B selects the externally actuated driver nodes, and u(t) is an input vector. A system is controllable if, for almost all nonzero edge weights compatible with the graph structure, there exists an input u(t) capable of moving the state from any initial condition to any final condition in finite time. This result is directly relevant to aging because the hallmarks of aging form an interacting directed network rather than a list of independent defects.

In the aging context, nodes may represent senescent-cell burden, mitochondrial dysfunction, proteostatic stress, stem-cell exhaustion, epigenetic drift, chronic inflammation, extracellular-matrix remodeling, nutrient-sensing activity, telomere dysfunction, genomic instability, altered intercellular communication, and impaired autophagy. Directed edges encode causal or regulatory dependencies: senescent cells increase inflammatory signaling; chronic inflammation increases damage; damage accelerates epigenetic drift; epigenetic drift impairs regenerative capacity; impaired regeneration increases fibrosis and functional decline. If this hallmark network were treated purely as a structural-control problem, the central question would be: what is the smallest set of biological nodes that must be directly actuated by interventions to render the aging network controllable?

This perspective clarifies why single-target geroprotectors may have limited restorative power. If the directed aging network has unmatched nodes distributed across multiple hallmark modules, then one driver node is insufficient for full structural controllability. A rapamycin-like intervention acting primarily through nutrient-sensing and proteostatic modules, for example, may alter the rate of decline but may not independently control senescent-cell burden, fibrosis, or epigenetic information loss. Conversely, a senolytic intervention may reduce inflammatory and paracrine senescence signaling but may not restore epigenetic state or regenerative potential. The network-control view therefore predicts that effective late-life rejuvenation will usually require multiple independent intervention vector fields.

Our framework extends the Liu–Slotine–Barabási approach in four ways.

First, we do not ask only whether the system is controllable in principle. We ask whether control is achievable at acceptable biological cost. Structural controllability ignores dose, toxicity, off-target effects, tissue-specific pharmacokinetics, and cancer risk. In the present framework, these enter through the running cost c(u), adverse-event penalty r(x, u), and viability constraints x(t) ∈ 𝒱. Thus, two intervention sets may be equivalent in structural controllability but radically different in clinical feasibility.

Second, we explicitly include safety constraints. Network controllability permits trajectories through biologically impossible or dangerous regions of state space. Aging interventions do not. Partial reprogramming may move epigenetic state toward youth while simultaneously increasing dedifferentiation or tumorigenic risk if applied outside a viability kernel. Senolysis may reduce senescent burden while impairing wound healing if excessive. Thus the relevant set is not the entire controllable set but the safe reachable set ℛ_safe defined in §4.3.

Third, aging is not a time-invariant control problem. The same drug vector field can have different effects in young, middle-aged, and old states because the biological substrate changes. For example, reprogramming efficacy depends on senescent burden, chromatin accessibility, immune surveillance, and regenerative competence. The matrix A, the admissible intervention matrix B, and the cost of actuation therefore change with age. This implies a progressive loss of controllability with time: some youthful states that are reachable from middle age may no longer be safely reachable from advanced age.

Fourth, our framework replaces the binary notion of controllability with a continuous value function. Biological age, in this formulation, is not merely distance from a young state but the minimum safe cost of restoration:

This definition incorporates both state deterioration and diminishing controllability. Two organisms with the same epigenetic-clock age may differ substantially in BA_control if one remains safely steerable toward a youthful functional state and the other does not.

The connection to resilience theory is also important. Gao, Barzel, and Barabási (2016) showed that diverse complex networks exhibit universal resilience patterns near critical transitions, allowing high-dimensional network dynamics to be reduced under some conditions to effective one-dimensional resilience coordinates. In aging, this suggests that some components of functional decline may reflect movement toward tipping points in a reduced resilience coordinate. Our framework is compatible with this reduction but adds control: the key question is not only how close the system is to a resilience threshold, but which interventions can move it away from that threshold safely and at what cost.

For a twelve-hallmark aging network, the number of independent drug vector fields required for practical control is unlikely to equal twelve. Strong coupling may permit control of several modules through one intervention. However, neither is it likely to be one. A plausible minimal intervention basis would include at least: an anti-senescent/senolytic field, a proteostasis/nutrient-sensing field, an epigenetic-reprogramming field, an anti-inflammatory field, a mitochondrial/metabolic field, and an extracellular-matrix/fibrosis field. Determining whether this basis is sufficient is an empirical network-identifiability problem, not a matter of verbal taxonomy. The contribution of the present control framework is therefore to translate hallmark biology into a testable question: which intervention vector fields span the safe, clinically relevant reachable set of aged organismal states?

5.Validation: The Insilico Pipeline as a Testbed

Aging theories often remain conceptually rich but translationally underconstrained. They explain phenomena but are rarely tested against the practical decisions of drug discovery: target nomination, medicinal chemistry, efficacy models, toxicology, preclinical candidate selection, and clinical readiness. A control-theoretic framework should be judged by whether it improves these decisions.

AI-driven drug discovery pipelines provide a natural validation environment. Since 2021, a corpus of approximately 30 AI-discovered preclinical candidate compounds can be treated as a retrospective testbed. Each program represents a decision sequence: target identification, disease selection, compound generation, optimization, efficacy testing, safety assessment, and advancement or termination. Even when proprietary details must remain blinded, the structure of the data permits quantitative comparison of target-scoring frameworks.

For each preclinical candidate program, one can compute multiple scores:

1. Hallmark score. Degree to which the target or pathway is annotated to canonical Hallmarks of Aging.
2. Age-association score. Evidence that the target changes with age across tissues, species, or omics datasets.
3. Disease-association score. Genetic, transcriptomic, proteomic, clinical, or mechanistic relevance to the target indication.
4. Network centrality score. Position in protein-protein interaction, regulatory, causal, or disease networks.
5. Druggability score. Structural tractability, ligandability, selectivity, and developability.
6. Toxicity-risk score. Predicted or observed liabilities from expression, essentiality, off-targets, and toxicology.
7. Control-value score. Estimated reduction in restoration cost V produced by modulating the target in the relevant biological state.

The central hypothesis is that the control-value score outperforms the Hallmark score for predicting translational advancement, functional efficacy, and toxicity-adjusted therapeutic index.

This comparison is important because Hallmark annotation may be necessary but not sufficient. Some successful targets may be strongly Hallmark-associated. Others may have low canonical Hallmark annotation but high control leverage because they regulate state transitions, tissue repair bottlenecks, fibrosis-inflammation coupling, stress recovery, or disease-aging interfaces. Conversely, some highly Hallmark-associated targets may fail because they do not move the relevant aged disease state toward functional viability under safety constraints.

A useful validation design would classify targets into four categories:

The control framework predicts A ≈ C > B > D. A Hallmark-driven model predicts A ≈ B > C ≈ D. This is a clear distinguishing test.

5.1 Retrospective endpoints

Possible retrospective endpoints include:

• nomination as a preclinical candidate compound;
• advancement through efficacy studies;
• reproducibility across disease models;
• magnitude of functional rescue;
• biomarker normalization;
• toxicology margin;
• pharmacokinetic and pharmacodynamic feasibility;
• progression toward IND-enabling studies;
• clinical readiness or clinical entry.

The analysis need not disclose proprietary structures or targets in full. Programs can be anonymized, with targets grouped by pathway class and disease area. The key question is whether control-value scoring explains advancement better than Hallmark annotation or static age association. Statistical evaluation could include logistic regression, survival analysis over development milestones, rank correlation with advancement stage, receiver operating characteristic analysis, precision-recall curves, and ablation studies. The primary endpoint should be pre-registered.

5.2 Rentosertib as an illustrative control-vector example

Rentosertib, a TNIK inhibitor discovered through an AI-enabled drug discovery process and developed for fibrotic disease, provides an illustrative example of how a target can be interpreted in control-theoretic rather than purely categorical terms. TNIK has been implicated in Wnt signaling, fibrosis-associated transcriptional programs, inflammatory remodeling, and senescence-fibrosis coupling. A conventional description might classify TNIK inhibition according to pathways or disease indication. A control-theoretic description asks a different question: does TNIK inhibition induce a vector field that moves the fibrotic aged tissue state toward a functional viability region?

In a fibrotic lung state, the relevant variables may include fibroblast activation, epithelial injury, extracellular matrix deposition, inflammatory signaling, senescent-cell burden, tissue stiffness, and gas exchange. TNIK inhibition may be valuable if its vector field reduces the restoration cost by altering the coupled senescence-fibrosis axis, improving tissue remodeling dynamics, or expanding the safe reachable set for subsequent interventions. Its value is not determined solely by whether TNIK belongs to a canonical Hallmark category. It is determined by whether modulating TNIK changes the trajectory of the diseased aged tissue in a function-restoring direction under safety constraints.

This example should not be treated as promotional evidence. Rather, it illustrates how real drug programs can be reinterpreted as empirical probes of controllability. Each compound tests whether a predicted vector field produces the desired state transition in biological systems and whether that transition remains safe.

5.3 Multi-model AI orchestration

The framework can incorporate multi-model AI orchestration at several stages: literature synthesis, hypothesis generation, target ranking, latent state inference, perturbation-response modeling, medicinal chemistry, and adversarial critique. Multiple AI systems can be used to generate candidate mechanisms, compare assumptions, identify missing evidence, and stress-test predictions. However, AI consensus should not be treated as evidence. It is a method for expanding and organizing hypothesis space. The evidentiary standard remains empirical validation.

A rigorous pipeline would separate: (1) AI-assisted hypothesis generation; (2) human-specified formal definitions; (3) pre-registered model comparison; (4) blinded retrospective validation; (5) independent prospective experiments. This distinction is essential. The novelty of the framework is not that AI systems can discuss aging theories. The novelty is the formal conversion of aging intervention into a safe-control problem and the possibility of testing that formalism against real drug discovery outcomes.

5.4 Limitations of the validation corpus

A single-company retrospective corpus has obvious limitations. It may reflect internal strategy, disease-area selection, platform biases, medicinal chemistry constraints, and unobserved decision criteria. The number of programs may be modest relative to the dimensionality of the models. Some failures may reflect pharmacokinetics or chemistry rather than target biology. Some successes may depend on indication-specific factors unrelated to aging.

Therefore, retrospective validation should be treated as an initial test, not final proof. Independent replication is required using public datasets, external drug discovery portfolios, animal intervention studies, clinical trial results, and prospective experiments. Nonetheless, such a corpus is valuable because it links theory to the practical realities of therapeutic development. Most aging theories are not tested against preclinical candidate nomination and toxicology. A control-theoretic framework can be.

5.5 The central empirical claim

The central empirical claim is not that all successful aging-related drugs will be Hallmark-independent, nor that Hallmark biology is unimportant. The claim is more precise:

If this claim fails, the framework loses practical value. If it succeeds, it provides an operational bridge from aging theory to drug discovery.

6.Worked Example: Optimal Senolytic–Reprogramming Control in Aged Murine Liver

To make the control-theoretic framework concrete, we introduce a deliberately low-dimensional, fully specified model of aged murine liver. The purpose of this example is not to claim that five variables capture all relevant hepatic aging biology. Rather, the purpose is to demonstrate how aging hallmarks, pharmacological interventions, safety constraints, non-commutativity, and value-function scoring can be placed into a single computable system.

6.1 State variables

Let

where:

• s is senescent-cell fraction;
• d is oxidative damage burden, interpretable as a normalized composite of 8-OHdG, protein carbonylation, and lipid peroxidation markers;
• e is epigenetic information integrity, with e = 1 denoting a youthful methylation state and e = 0 denoting maximal age-associated drift;
• r is regenerative capacity, summarizing hepatocyte proliferative competence, progenitor function, and tissue repair ability;
• f is fibrotic burden, interpretable as normalized collagen deposition or extracellular-matrix remodeling.

The uncontrolled aged state is initialized as a 24-month-old mouse liver:

These values are not asserted to be universal constants. They are normalized representative values chosen to match the qualitative range reported in aged murine tissues: increased senescent-cell markers, increased oxidative damage, reduced methylation-clock youthfulness, reduced regenerative potential, and increased fibrotic burden.

6.2 Controlled dynamics

We use the following stochastic controlled system (shown here in component form; in SDE notation the drift vector is b(x, u) and the noise matrix is Σ):

Equivalently, in stochastic differential-equation notation,

with diagonal noise matrix

For the deterministic simulations below we report the mean trajectory, i.e. ξ_i(t) = 0. Stochastic simulations with these noise amplitudes preserve the same protocol ranking.

The intervention variables are:

• u_sen(t), senolytic exposure;
• u_rep(t), cyclic partial-reprogramming exposure;
• u_rap(t), rapamycin exposure;
• u_tnik(t), TNIK-inhibitor exposure.

For the senolytic–reprogramming example, u_rap = 0 and u_tnik = 0. They are included to show how the same model accommodates additional gerotherapeutic vector fields.

Because senolytics and reprogramming are typically administered in courses or pulses, the control functions are normalized so that

Thus β_s is interpreted as senescent-cell clearance per senolytic course, and μ_e as epigenetic restoration per reprogramming pulse.

6.3 Parameterization

The parameter values used in the worked example are listed below. They should be read as literature-calibrated illustrative values rather than definitive hepatic-aging constants.

The reprogramming parameter μ_e = 0.150 per pulse is calibrated to the order of magnitude of partial epigenetic reversal reported by Ocampo et al. (2016) and Lu et al. (2020). Ocampo et al. showed that cyclic OSKM induction ameliorated age-associated phenotypes in progeroid mice and improved tissue function without continuous dedifferentiation when delivered cyclically. Lu et al. reported approximately 30% methylation-age reversal in retinal ganglion cells using OSK-mediated epigenetic reprogramming. We therefore set two effective pulses to produce approximately 30% reversal in a permissive low-senescence, high-regeneration state.

The senolytic parameter β_s = 0.800 per course corresponds to a one-course reduction factor exp(−0.8) ≈ 0.45, i.e. approximately 55% clearance of the susceptible senescent-cell compartment. This lies within the range of preclinical senolytic effects reported for targeted removal of p16-positive cells and senescent-cell-enriched populations. Baker et al. (2011) demonstrated that clearance of p16^Ink4a-positive senescent cells delayed age-associated disorders and extended median lifespan in BubR1 progeroid mice by approximately 20–30%, motivating the inclusion of a high-efficacy senolytic field.

The TNIK parameter is calibrated to reported 30–50% reductions in fibrosis-associated markers after TNIK inhibition. Rapamycin is included because Harrison et al. (2009) showed that late-life rapamycin increased murine lifespan by approximately 14%, supporting the representation of mTOR inhibition as a damage- and stress-modifying vector field rather than as a complete rejuvenation operator.

6.4 Viability set and protocols

For this example, the viability set is

These constraints encode avoidance of excessive senescent burden, severe damage, epigenetic collapse, regenerative failure, and advanced fibrosis. During numerical integration, trajectories exiting 𝒱 incur a large penalty; admissible protocols remain inside 𝒱.

We compare three fixed protocols over T = 56 days:

The model was integrated using explicit Euler time stepping with Δt = 0.25 days. State variables were projected to [0, 1] after each step. The reported values are deterministic mean trajectories.

6.5 Simulation results and value-function comparison

The final states at day 56 are:

Thus, in this parameterized example, senolysis before reprogramming produces greater restoration of epigenetic integrity:

The reason is mechanistic. Reprogramming efficacy is proportional to (1 − s) r. If senescent burden is high, the reprogramming vector field is attenuated. Clearing senescent cells first increases the effective gain of the reprogramming pulse. If reprogramming is applied before senolysis, part of the pulse is wasted in a high-senescence, low-regenerative-permissiveness state.

For each protocol u, define the finite-horizon cost

where P_𝒱 = 10³ if the trajectory exits 𝒱 and 0 otherwise. The terminal cost is

For the numerical example, we set (q_s, q_d, q_e, q_r, q_f) = (1, 1, 2, 1, 1), (w_s, w_d, w_e, w_r, w_f) = (2, 1, 4, 2, 1), and intervention costs c_s = 0.010, c_r = 0.015. The resulting restricted protocol costs are:

Therefore, over this restricted protocol class,

This is not the exact Hamilton–Jacobi–Bellman value over all admissible policies. It is a finite-protocol approximation demonstrating how candidate intervention schedules can be scored by the same value-function logic. A full optimal-control solution would optimize over pulse timing, dose, duration, and combinations, subject to viability constraints.

6.6 Lie-bracket demonstration of non-commutativity

The above ordering effect can be shown analytically. With state ordering x = (s, d, e, r, f), the senolytic controlled vector field is

and the reprogramming vector field is

The Lie bracket is

The Jacobian of g_rep has nonzero entries

The Jacobian of g_sen has one nonzero entry,

Therefore,

whereas Dg_sen(x) g_rep(x) = 0, because g_rep has no s-component. Hence

At the aged initial state,

The nonzero bracket means that the flow generated by senolysis followed by reprogramming is not equivalent to the flow generated by reprogramming followed by senolysis. Biologically, the bracket term is positive in the epigenetic-integrity coordinate because senolysis increases the effective gain of reprogramming by reducing the inhibitory factor s in μ_e(1 − s) r. This is the mathematical expression of the ordering effect observed in the simulation.

The worked example makes four points. First, even a small aging-state model can generate intervention-order effects that are invisible in static hallmark diagrams. Second, the superiority of Protocol A is not assumed; it follows from the state dependence of the reprogramming vector field. Third, the value-function score provides an operational way to compare schedules by integrating endpoint restoration, cumulative burden, intervention cost, and safety. Fourth, the model yields quantitative kill criteria: if, in aged murine liver, senolytic pretreatment does not reduce senescent-cell burden and does not increase the epigenetic gain of subsequent OSK-like reprogramming, then the proposed non-commutative mechanism is rejected for this tissue and intervention pair.

7.Twenty Novel Predictions

A useful theory of aging drug discovery must do more than reorganize known mechanisms; it must generate predictions that would not follow from a static catalog of aging phenotypes. The Hallmarks framework identifies recurrent biological processes associated with aging (López-Otín et al., 2013; 2023). However, a hallmark list does not specify the geometry of the aging state space, the order-dependence of interventions, the controllability of different organismal states, the energetic or safety cost of moving between states, or the conditions under which an intervention becomes futile.

The control-theoretic framework introduced here does. It treats aging as a trajectory x(t) through a high-dimensional biological state space, with intrinsic drift f(x), intervention vector fields g_i(x), a viability region 𝒱, and a value or cost function V(x) representing biological age, loss of controllability, or distance-to-failure. This immediately generates experimentally testable predictions that cannot be derived from Hallmarks alone.

Together, these predictions illustrate the principal advantage of a control-theoretic theory of aging: it is not merely descriptive. It makes quantitative, falsifiable claims about order, timing, state dependence, safety, synergy, irreversibility, tissue specificity, sex differences, and translational probability. Many of these predictions may prove wrong in detail. That is not a weakness but a requirement for progress. A theory that cannot fail experimentally cannot guide drug discovery.

8.Relationship to Existing Frameworks

The control-theoretic framework is not proposed as a replacement for existing theories of aging. Rather, it provides a mathematical language in which several major frameworks can be expressed, compared, and operationalized for intervention design.

8.1 Hallmarks of Aging

The Hallmarks framework organizes diverse aging mechanisms into a finite set of recurrent biological processes (López-Otín et al., 2013; 2023). In the present framework, hallmarks correspond to state variables, latent coordinates, or coarse-grained axes of the state vector x. A state-space model can include variables representing DNA damage burden, chromatin state, mitochondrial performance, senescent-cell abundance, inflammatory tone, immune repertoire diversity, extracellular matrix stiffness, metabolic flexibility, and stem-cell reserve. Thus, Hallmarks biology provides much of the biological vocabulary from which the state space is constructed.

However, Hallmarks alone does not specify dynamics. It does not define the drift f(x), intervention vector fields g_i(x), viability boundaries, control costs, reachable sets, or optimal policies. The control-theoretic framework adds these missing elements and, as shown in §4.6, translates hallmarks into a testable network-identifiability question: which intervention vector fields span the safe reachable set?

8.2 SENS and Damage Repair

The SENS framework emphasizes periodic repair or removal of categories of age-associated damage (de Grey et al., 2002; de Grey, 2007). The control framework formalizes SENS-like repair as a set of damage-clearance vector fields. A senolytic reduces the senescent-cell coordinate and secondarily changes inflammatory and tissue-remodeling coordinates; a crosslink-breaker alters ECM coordinates; mitochondrial interventions alter bioenergetic and redox coordinates.

The added contribution is state dependence. The same repair intervention may have different effects depending on the current state x. Clearing senescent cells in a mildly aged tissue may be beneficial; clearing too aggressively in a frail tissue could impair wound repair or destabilize a compensatory inflammatory equilibrium. Thus, the control formulation preserves the repair logic of SENS while embedding it in a dynamic model of dose, timing, tissue context, and safety.

8.3 Sinclair Information Theory of Aging

The information theory of aging emphasizes loss of epigenetic information and proposes that restoration of youthful gene-regulatory information can reverse aspects of aging (Ocampo et al., 2016; Lu et al., 2020; Sinclair and LaPlante, 2019). In control-theoretic terms, information restoration is a powerful but dangerous control direction. Reprogramming vector fields can move chromatin and transcriptional coordinates toward youthful regions, but they may also move cell-identity coordinates toward unsafe regions if applied excessively. Thus, reprogramming has both high control authority and high risk. The central problem is not merely whether epigenetic information can be restored, but how much, when, in which tissue, in what order relative to other interventions, and with what feedback constraints.

8.4 Geroscience and Hyperfunction Theory

The geroscience hypothesis proposes that targeting fundamental mechanisms of aging can delay or prevent multiple chronic diseases simultaneously (Kennedy et al., 2014; Barzilai et al., 2016; 2018). The control framework provides the formal optimization language that geroscience requires. Clinical trial design should shift from chronological-age enrollment to biological-state stratification, from fixed dosing to adaptive dosing, and from target engagement alone to verified movement of the system toward lower V(x).

Hyperfunction theory (Blagosklonny, 2006; 2013) identifies persistent growth signaling as a driver of late-life pathology. Within our framework, mTOR-driven hyperfunction is a component of the intrinsic drift field f(x); rapamycin is an opposing control field g_rapa(x) that can slow or redirect this drift. However, control theory predicts that rapamycin efficacy depends on state, timing, dose, tissue, sex, and coupling. It may be highly effective when hyperfunction is a dominant drift component and less effective when irreversible tissue damage, fibrosis, clonal hematopoiesis, or immune exhaustion dominate the state.

8.5 Identifiability, Model Risk, and Robust Control

The framework proposed here is useful only to the extent that its state variables, parameters, and intervention vector fields can be estimated. A formal control model without identifiability is not an operational model. We therefore distinguish structural identifiability, practical identifiability, and model risk.

Structural identifiability asks whether parameters are uniquely recoverable from perfect, noise-free observations of the model outputs. In systems biology, this question is often nontrivial even for modest ODE models. Raue et al. (2009) emphasized that many biological models contain parameter combinations that produce indistinguishable trajectories, making individual parameters unidentifiable without additional measurements or perturbations. Practical identifiability asks the more stringent experimental question: given finite sampling, measurement noise, biological heterogeneity, and limited perturbation regimes, can the parameters be estimated with sufficiently narrow confidence intervals to support prediction and control?

For the five-dimensional liver model in §6, some parameters are more identifiable than others. Longitudinal measurements of senescence markers (p16^Ink4a, p21, SA-β-gal, SASP markers) before and after senolytic treatment can identify β_s more directly than α_s, because the intervention produces a sharp perturbation in s. Similarly, methylation-clock or chromatin-state measurements before and after defined OSK pulses can identify the product μ_e(1 − s) r, but not necessarily μ_e, s, and r separately unless senescent burden and regenerative capacity are measured independently. Fibrosis-marker measurements before and after TNIK inhibition can help identify ψ_f, whereas the spontaneous fibrosis-generation parameter φ_f requires longer untreated or control trajectories.

Several parameters are identifiable only as combinations under typical experimental designs. For example, in

short-horizon reprogramming experiments identify the net reprogramming gain μ_e(1 − s) r more readily than μ_e alone. In

ν_r and ρ_r may trade off unless regenerative capacity is measured under perturbations that independently vary damage, senescence, fibrosis, and epigenetic integrity. Thus, multi-omics observation alone is insufficient; informative perturbations are required.

A feasible identifiability design for this model would include longitudinal multi-omics and histological measurements at baseline and multiple post-intervention time points under at least four experimental arms: control, senolytic alone, reprogramming alone, and senolytic followed by reprogramming. Additional rapamycin and TNIK-inhibitor arms would be needed to identify η_d and ψ_f. Outputs should include senescence burden, oxidative-damage markers, methylation-clock estimates, regenerative assays, collagen/fibrosis quantification, and adverse-event markers. Parameter inference could then proceed through profile likelihood, Bayesian posterior sampling, or sequential Monte Carlo, with explicit reporting of confidence or credible intervals.

Even if identifiable, the model may be wrong. Model misspecification is a central risk in biological control. The true system may contain omitted variables such as immune surveillance, clonal expansion, hepatic zonation, metabolic state, sex differences, microbiome effects, or cancer risk. Functional forms may be wrong: reprogramming may not scale as (1 − s) r; senolytic efficacy may saturate; fibrosis may include irreversible components; epigenetic restoration may decouple from functional restoration. Under such misspecification, an HJB-optimal policy for the assumed model can be unsafe or ineffective in the real system.

Robust control is therefore the appropriate mathematical setting for translation. Instead of optimizing a single nominal model, one may define an uncertainty set ℳ of plausible models and solve a minimax problem:

This formulation penalizes policies that work only under optimistic assumptions. H_∞-control methods provide one classical approach when uncertainty can be represented as bounded disturbances. Distributionally robust optimization provides another when uncertainty is represented as a family of probability distributions around empirical estimates. In adaptive trials, model-predictive control (MPC) is likely more realistic than solving a full HJB equation: the policy is repeatedly updated as new biomarker and safety data arrive.

8.6 A Shared Mathematical Language

The unifying claim is modest but important: the control-theoretic framework supplies a mathematical language that existing aging theories need. Hallmarks identify coordinates; SENS identifies repair directions; information theory emphasizes epigenetic state restoration; geroscience defines the translational objective; hyperfunction theory specifies an important drift component. Control theory integrates these elements into a computable system with state variables, dynamics, interventions, cost functions, constraints, and feedback. The central question of drug discovery then becomes not "Which hallmark does this target affect?" but "How does this intervention move the current biological state, with what controllability, safety, durability, and opportunity cost?"

9.Limitations

The framework proposed here is intentionally ambitious and should be interpreted as a research program rather than a completed model. Several limitations must be acknowledged.

First, the biological state space is extremely high-dimensional. Aging involves molecular, cellular, tissue, systemic, behavioral, environmental, and clinical variables. Direct modeling of this full space is impossible with current data. Dimensionality reduction, latent-state modeling, causal representation learning, and mechanistic coarse-graining will be required, but these introduce assumptions and may discard important variables.

Second, parameter identifiability remains a major challenge, as discussed formally in §8.5. Most existing datasets are cross-sectional. Even longitudinal data are often sampled infrequently, with incomplete omic coverage and uncontrolled interventions. Inferring causal dynamics from such data is difficult and may produce non-identifiable models.

Third, tissue specificity complicates organism-level modeling. Biological age is not a scalar property uniformly shared across tissues (Horvath, 2013; Zhang et al., 2019). Organism-level outcomes emerge from interactions among tissue-specific states. A drug that improves one organ may harm another. A clinically useful model must therefore integrate tissue-specific controllability maps with whole-organism cost functions.

Fourth, stochasticity and model uncertainty are not optional details. Aging trajectories are shaped by random mutations, infections, injuries, inflammatory events, environmental exposures, and clonal expansions. Two organisms with similar baseline biological age may diverge substantially. Deterministic optimal-control solutions may be misleading if they ignore stochastic transitions, unobserved variables, and measurement error. Robust control (§8.5) is therefore a prerequisite for translation.

Fifth, computational tractability is a serious obstacle. Optimal control in high-dimensional, nonlinear, partially observed systems is mathematically and computationally difficult. Exact solutions will rarely be available. Practical applications will require approximations such as model predictive control, reinforcement learning, neural ordinary differential equations, causal state-space models, and constrained optimization. These methods can fail, overfit, or propose unsafe policies if not constrained by biological knowledge and experimental validation.

Sixth, retrospective validation using drug-discovery pipelines is not prospective proof. Retrospective analyses are vulnerable to selection bias, publication bias, incomplete negative data, and post-hoc parameter tuning. Prospective experiments are essential.

Seventh, the required longitudinal multi-omics datasets do not yet exist at sufficient scale. The framework demands repeated measurement of biological state before, during, and after intervention, ideally across tissues, sexes, ages, environments, and genotypes. It also requires standardized perturbation maps for drugs and combinations.

Finally, biological-age cost functions themselves remain imperfect. Epigenetic clocks, proteomic clocks, metabolomic clocks, frailty indices, and clinical risk scores each capture partial information (Horvath, 2013; Levine et al., 2018; Lu et al., 2019; Pyrkov and Fedichev, 2019). None is equivalent to "true aging." The value function V(x) must be treated as an evolving construct that improves as better outcomes, biomarkers, and mechanistic data become available.

These limitations do not invalidate the framework. They define the agenda. The appropriate comparison is not between a perfect control-theoretic model and imperfect biology, but between an explicit, testable, improvable model and informal reasoning that cannot make quantitative predictions.

10.Discussion

Aging drug discovery is entering a new phase. The field has moved beyond the question of whether aging biology is modifiable. Dietary restriction, rapamycin, genetic perturbations, senolytic strategies, partial reprogramming, exercise, and other interventions demonstrate that biological aging trajectories can be altered in model systems (Harrison et al., 2009; Fontana et al., 2010; Baker et al., 2011; 2016; Ocampo et al., 2016; Lu et al., 2020). The central question is now how to design interventions that are effective, safe, durable, translatable, and personalized.

The control-theoretic framework offers five contributions. First, it defines aging as movement through a biological state space rather than as a list of independent mechanisms. Second, it treats drugs as vector fields whose effects depend on the current state. Third, it introduces controllability: the ability of feasible interventions to move an organism toward healthier regions. Fourth, it formalizes biological age as a cost or value function related to future risk and intervention difficulty. Fifth, it makes falsifiable predictions about order, timing, synergy, irreversibility, tissue specificity, and adaptive treatment.

This approach is particularly important because gerotherapeutics are unlikely to behave like conventional single-disease drugs. A cancer drug may be judged by tumor response; an antihypertensive by blood pressure; an antibiotic by pathogen clearance. A gerotherapeutic must alter the future probability of multiple diseases, functional decline, frailty, and mortality, often over long timescales. Its effect may depend strongly on baseline state. Its risks may arise from overshooting youthful pathways or destabilizing compensatory adaptations. These are control problems.

A Five-Year Experimental Agenda

A realistic five-year agenda should prioritize tractable experiments that test framework-specific predictions.

First, the field should construct perturbational vector-field atlases for major gerotherapeutic classes. Aged primary cells, organoids, and mice should be exposed to rapamycin, senolytics, metformin-related compounds, NAD modulators, anti-inflammatory agents, mitochondrial interventions, autophagy enhancers, matrix remodelers, and partial reprogramming regimens. Multi-omic measurements before and after perturbation would estimate g_i(x) across states.

Second, order-dependence experiments should be performed. Senolytic–reprogramming, rapamycin–senolytic, matrix-remodeling–immune-reset, and other combinations should be tested in systematic permutations. These experiments directly test non-commutativity (§6.6) and will reveal whether intervention scheduling is a major source of efficacy variation.

Third, biological-state stratification should replace chronological-age stratification wherever possible. Mouse studies should enroll animals by methylation age, proteomic age, frailty, immune state, and organ-specific function. Human trials should similarly incorporate biological-age and resilience biomarkers. This will determine whether controllability windows exist.

Fourth, adaptive-control trials should be launched in mice. Fixed-dose rapamycin or senolytic regimens should be compared with biomarker-guided dosing using methylation feedback, inflammatory markers, weight, glucose, immune status, and frailty. If feedback control improves efficacy or safety, it will justify more sophisticated adaptive clinical designs — ideally formulated as model predictive control with robust-control guarantees (§8.5).

Fifth, longitudinal multi-omic cohorts should be used to identify tipping points. The goal is to detect early-warning signatures of irreversible controllability loss before clinical frailty or disease appears. Such signatures would be valuable both for prevention and for trial enrichment.

Digital Twins and Personalized Gerotherapy

The long-term vision is a digital twin for aging: a personalized computational model that estimates an individual's biological state, forecasts likely trajectories, simulates candidate interventions, and recommends an optimal policy under safety constraints. This vision is not science fiction, but it requires disciplined development. A digital twin must be grounded in longitudinal data, calibrated against perturbational responses, and continuously updated with feedback.

In such a system, a patient's methylome, proteome, metabolome, immune profile, microbiome, clinical history, imaging, functional measures, and wearable data would define an estimated state x̂(t). The model would compute a value function V(x̂), estimate reachable healthier states, and compare intervention policies. The output would not be a generic recommendation such as "take rapamycin" but a constrained policy: dose, timing, monitoring, stopping criteria, combination logic, and expected uncertainty.

Personalization is essential because aging is heterogeneous. Two individuals of the same chronological age may differ in immune aging, vascular stiffness, epigenetic entropy, kidney function, sarcopenia, senescent-cell burden, and inflammatory tone. A fixed gerotherapy protocol may help one, harm another, and do nothing for a third. Control theory provides the mathematical foundation for individualized intervention.

Why AI Is Essential

Artificial intelligence is not optional in this framework. The state space is too large, the data too heterogeneous, and the control problem too complex for manual reasoning alone. AI is required at several levels.

First, AI can estimate biological state from incomplete, noisy data. Multimodal models can integrate methylation, transcriptomics, proteomics, metabolomics, imaging, wearables, and clinical records into latent representations relevant to aging.

Second, AI can infer intervention vector fields from perturbational data. Large-scale drug-response maps can reveal how compounds move biological states and whether those movements are favorable, redundant, or unsafe.

Third, AI can optimize control policies. Reinforcement learning, model predictive control, Bayesian optimization, and causal inference can help identify dosing schedules and combinations that minimize V(x) under constraints.

Fourth, AI can design new drugs. If the desired control direction is known, generative chemistry and target-discovery systems can search for molecules or biologics that approximate that vector field with acceptable safety and pharmacology.

Fifth, AI can support trial design. It can identify patients or animals near controllability windows, predict responders, monitor divergence from expected trajectories, and adapt protocols in real time.

However, AI must be constrained by biology and validation. Predictive performance on retrospective data is insufficient. Models must make prospective, falsifiable predictions and be tested in controlled systems. The goal is not to replace experimental geroscience, but to make it more quantitative and efficient.

From Measurement to Prescription

The ultimate translational pathway can be summarized in four steps. First, measure biological state. This requires robust, affordable, repeatable assays that capture relevant aging dimensions across tissues and systems. Second, estimate risk and controllability. A biological-age score alone is insufficient; the model must estimate whether the state remains modifiable and by which interventions. Third, simulate intervention policies. Candidate drugs, combinations, doses, and schedules should be compared for expected V(x)-reduction, safety, durability, and uncertainty. Fourth, prescribe and adapt. Treatment should be monitored with feedback, updated as the state changes, and stopped or modified if the trajectory becomes unsafe.

This vision reframes aging medicine. Instead of treating diseases after they emerge, clinicians would manage biological-state trajectories before irreversible loss of function occurs. Instead of asking whether a patient is "old," the relevant questions would be: Where is the patient in state space? Which failure modes are approaching? Which controls remain effective? What policy reduces future risk with acceptable cost?

The control-theoretic framework is therefore both a scientific theory and an engineering program. Its success will depend not on rhetorical appeal, but on whether it improves prediction, experiment design, drug discovery, and clinical outcomes. The twenty predictions above provide a starting point.

11.Materials, Methods, and AI Disclosure

This manuscript was developed as a conceptual synthesis integrating aging biology, control theory, geroscience, and AI-enabled drug discovery. The framework was constructed through iterative abstraction from established aging mechanisms, intervention classes, biomarker systems, and dynamical-systems principles.

The worked example of §6 was implemented as a five-dimensional controlled stochastic differential equation and integrated by explicit Euler stepping with Δt = 0.25 days over T = 56 days. State variables were projected to [0, 1] after each step. Parameter values are literature-calibrated illustrative values (see Table in §6.3); they are not presented as definitive hepatic-aging constants. The reported final states are deterministic mean trajectories; stochastic simulations with the stated noise amplitudes preserved the same protocol ranking. The restricted value-function comparison uses the weights specified in §6.5 and a viability penalty P_𝒱 = 10³. Full code and parameter files are available from the authors on request.

The Lie-bracket calculation of §6.6 is analytical and can be verified by hand from the vector-field definitions. The non-zero bracket in the epigenetic coordinate reflects the multiplicative (1 − s) factor in the reprogramming vector field.

No new wet-laboratory experiments, animal studies, or human-subject data are presented. Instead, the paper proposes a theoretical and computational framework and identifies prospective experiments capable of falsifying or refining the framework. Where specific interventions are discussed, they are used as examples of possible control vector fields and not as clinical recommendations.

Revised AI Disclosure

Artificial intelligence tools were used as assistive instruments during manuscript preparation but were not authors and were not assigned responsibility for scientific content. In accordance with ICMJE guidance, authorship is restricted to human contributors who made substantial scholarly contributions, approved the final manuscript, and accept accountability for the integrity of the work.

AI systems were used for limited drafting support, organization of reviewer-response materials, language editing, mathematical-format checking, and generation of alternative phrasings during revision. The workflow was iterative: human authors supplied prompts specifying the scientific content to be developed; AI tools generated draft text or structural suggestions; human authors then reviewed, corrected, rejected, or rewrote the output. No AI-generated statement was accepted without human review.

All factual claims, citations, mathematical definitions, parameter interpretations, and biological assertions were checked by the human authors against primary literature or established secondary sources. Where the manuscript contains modeling assumptions, parameter choices, or illustrative simulations, these are explicitly identified as such and remain the responsibility of the human authors. AI tools were not used to fabricate data, conduct independent experiments, or generate undisclosed empirical results.

The authors recognize that large language models can introduce systematic errors, including inaccurate citations, overconfident summaries, bias toward highly cited literature, and omission of negative or contradictory evidence. For this reason, AI-generated bibliographic or factual material was treated as provisional until independently verified. The final manuscript, including its claims, limitations, equations, and interpretations, was written, checked, and approved by the human authors, who accept full responsibility for its accuracy and integrity.

12.Conclusion

Aging biology has generated powerful descriptive frameworks, but drug discovery requires more than description. It requires a theory that can predict which intervention should be applied, when, in whom, in what order, at what dose, and with what monitoring. We propose that control theory provides this missing language.

In this framework, aging is a trajectory through biological state space; disease and frailty arise as the system approaches or exits a viability region; drugs are control vector fields; biological age is a cost or value function; and gerotherapy is an optimal-control problem under uncertainty and safety constraints. This formulation naturally explains why intervention effects are state-dependent, why order can matter, why combinations may be safer than high-dose monotherapy, why tissues differ in reversibility, and why some states become effectively irreversible. The worked example of §6 makes these claims quantitative in a fully specified five-dimensional model of aged murine liver, with an explicit Lie-bracket calculation showing that the senolytic and reprogramming vector fields do not commute.

The framework does not replace Hallmarks, SENS, information theory, geroscience, or hyperfunction theory. Instead, it provides a mathematical structure in which each can be represented and connected. Hallmarks define coordinates; SENS defines repair vectors; reprogramming defines an information-restoration vector; hyperfunction defines part of the aging drift; geroscience defines the translational objective. Network controllability theory (§4.6) and systems pharmacology (§4.5) provide the broader mathematical and empirical context within which the gerotherapeutic control problem is formulated.

Most importantly, the framework is falsifiable. The twenty predictions outlined in §7 can be tested in mice, organoids, perturbational omics, retrospective clinical datasets, and prospective trials. Some will fail. Those failures will refine the model. Identifiability and model-risk analysis (§8.5) make the conditions for that refinement explicit.

The long-term goal is personalized aging control: measure biological state, estimate controllability, simulate interventions, prescribe an optimal policy, and update treatment through feedback. Achieving this will require longitudinal multi-omics, perturbational atlases, AI-enabled state estimation, robust-control formulations, and rigorous experimental validation. If successful, aging drug discovery can move from hallmark targeting to rational control of biological trajectories.