A stochastic agent-based extension of the GSM squared model for particle therapy: cell-cycle dynamics, dose-rate dependence, and fractionation effects

arXiv:2604.27630 · physics.bio-ph · Submitted 2026-04-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "A stochastic agent-based extension of the GSM squared model for particle therapy".

Mira: Here's the gist of the paper. Accurately linking microscopic energy deposition from ionizing radiation to emergent biological outcomes remains a central challenge in radiobiological modeling, particularly when stochastic damage induction, cell-cycle dynamics, and spatial organization within irradiated tissues must be treated explicitly and consistently across scales<ref:2604.27630#pg2> The framework is applied to three-dimensional tumor spheroids irradiated with 1H and 12C ions across a range of energies and dose rates<ref:2604.27630#pg10> Several empirically established trends in biological response, including the dose-rate dependence of cell survival, its attenuation at high LET, and the inverse dose rate effect in split-dose irradiation, emerge from the model through the explicit coupling of particle arrivals, damage accumulation, and repair kinetics<ref:2604.27630#pg10> The model captures repair, reassortment, and repopulation within a single consistent framework and provides a basis for quantifying stochastic variability in treatment response at the single-cell level<ref:2604.27630#pg10>.

Kai: Modeling Heterogeneity through Agent-Based Dynamics The researchers developed an agent-based modeling framework that extends the GSM2 to explicitly resolve individual cells and their biological dynamics following particle irradiation<ref:2604.27630#pg9> Each cell is represented as an autonomous agent whose internal state evolves according to a continuous-time Markov chain (CTMC)<ref:2604.27630#pg9> System evolution is simulated using a next-event, event-driven algorithm, enabling efficient computation at single-cell resolution over large spatial domains.

Mira: Incorporating Physical and Biological Scales The framework integrates several physical and biological components to capture the complexity of radiation response<ref:2604.27630#pg9> The model explicitly includes: 1.

Kai: Radiation field modeling, where particle arrivals are sampled stochastically from an exponential distribution with a rate proportional to the prescribed dose rate.

Title and authors: Kai: Let's talk about who wrote this. It’s Francesco G. Cordoni, Marco Battestini, and Marta Missiaggia from Trento Institute for Fundamental Physics and Application at TIFPA, along with some folks from LSU in the US.

Mira: They’ve put a lot of work into extending the GSM2 model here to make it fully stochastic and agent-based. It really shows how they are tackling those fundamental modeling challenges head-on.

Lev: From my side, I'm thinking about how much overhead this kind of explicit simulation adds when you try to run it on actual hardware, especially since we are dealing with next event algorithms and continuous time Markov chains here.

Kai: It’s a lot of computation, but the goal is getting that single-cell resolution to capture the biological dynamics accurately. They use an event-driven simulation algorithm which lets them be efficient over large spatial domains.

Mira: And they’ve built in a way to handle four different types of events: repair and lethal lesion formation, cell cycle transitions, irradiation particle arrivals, and cell migration. It’s a comprehensive approach to the complexity.

Lev: That event-driven nature is what makes it computationally feasible for this level of detail; it avoids those slow fixed time steps when you are tracking individual cells evolving their state according to a continuous time Markov chain.

Kai: So, they’ve connected the physical process—the particle hitting the cell—directly to the biological response in a way that is much more granular than what population averages usually give us.

Mira: And this connects directly to their next goal: predicting TCP and NTCP, which means linking these microscopic events to macroscopic clinical outcomes.

The paper's summary: Kai: So the core of this paper, "A stochastic agent-based extension of the GSM squared model for particle therapy: cell-cycle dynamics, dose-rate dependence, and fractionation effects," is that they are extending the GSM2 model to be a stochastic agent-based radiobiological modeling framework.

Mira: They are specifically looking at how cell cycle dynamics, dose rate dependence, and fractionation effects interact within this single consistent framework. It’s about moving beyond simple deterministic models to capture the true biological noise.

Lev: What I see is they are trying to resolve the multi-layered heterogeneity that exists in a tumor—the physical environment, the cell cycle state, and the random damage induction at every single point.

Kai: They integrate several physical and biological components: radiation field modeling where particle arrivals are sampled stochastically based on dose rate, oxygenation modeled through a reaction diffusion equation for spatial variation, and DNA damage induction based on linear energy transfer and oxygenation via the Oxygen Enhancement Ratio.

Mira: The key contribution here is how they resolve the cell cycle phase variations spatially, from a quiescent core to a cycling shell around it. They show that this spatial organization evolves dynamically after irradiation, which reflects how damage, repair kinetics, and cell cycle progression interact in space.

Lev: That spatial resolution is what makes it so much more powerful than just looking at a uniform dose distribution; you can see how the environment dictates the survival probability for different cells.

Kai: And they show that because of this explicit representation, they reproduce three classical radiobiological rates: repair, reassortment, and repopulation all in one place.

The paper's improvements: Mira: Now let's talk about what they suggest as improvements. They are looking at how the model can be refined to capture more of the biological reality.

Kai: One limitation they note is that cell fate is treated as binary, meaning it doesn't explicitly account for death modalities like apoptosis or radiation-induced senescence. That’s one thing they acknowledge they haven't included yet.

Lev: That’s a fair critique; most of these models simplify the endpoint to just alive or dead, but those other processes matter in reality. I’d say that would add significant complexity to the Markov chain transition rates you mentioned earlier.

Mira: Another point is oxygenation, which they model as a quasi-static radial profile without dynamic coupling to actual vascular supply or nutrient transport. They flag this as something left for future development because it’s a major source of heterogeneity.

Kai: So, the paper itself admits that modeling dynamic reoxygenation and explicit vascular descriptions is next on the list for future work, which makes sense given the complexity of tissue response.

Lev: From an engineering standpoint, that missing dynamic coupling would require a whole new layer of simulation complexity because you’d need to model fluid dynamics alongside the cellular state.

Mira: But even without it, they show how their current structure reproduces dose-rate effects without needing empirical correction factors like the Lea–Catcheside factor.

Conclusion: Kai: To wrap up on this paper, the main implication is that this agent-based extension of GSM2 provides a mechanistic, cell-population radiobiological model that accounts for repair, reassortment, and repopulation in a single consistent framework.

Mira: It generates distributions of possible outcomes rather than just single deterministic predictions. That dual stochasticity allows them to quantify uncertainty in treatment response under realistic irradiation conditions.

Lev: For me, the most important thing is that this framework provides a mechanistic link between microscopic damage processes and macroscopic clinical predictions like TCP and NTCP by aggregating single-cell outcomes into population metrics.

Kai: So, the full title, "A stochastic agent-based extension of the GSM squared model for particle therapy: cell-cycle dynamics, dose-rate dependence, and fractionation effects," gives us a tool that lets us see exactly how things happen at every level of complexity.

Mira: It’s a powerful way to understand why certain treatment responses are observed without needing to adjust parameters empirically just to fit the data.

Lev: Yeah, it’s a step forward in understanding the physics of biological response under these kinds of complex irradiation scenarios.

Trento Institute for Fundamental Physics and Application · Department of Civil, Environmental and Mechanical Engineering University of Trento · Department of Civil and Environmental Engineering Louisiana State University Department of Physics & Astronomy Louisiana State University

physics.bio-ph

Submitted: 2026-04-30

Updated: 2026-11-09

Code: https://github.com/francescogcordoni/GSM2_Julia

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 80/100

The gist: Accurately linking microscopic energy deposition from ionizing radiation to emergent biological outcomes remains a central challenge in radiobiological modeling…

Key concepts

Agent-Based Modeling (ABM)
This approach treats every cell as an individual agent with its own internal state that changes over time according to defined rules. This allows the simulation to capture cell-to-cell variability in how they react to radiation and how they evolve through different biological stages, moving beyond simple population averages.
Continuous-Time Markov Chain (CTMC)
The internal state of each cell is modeled using a CTMC, which describes the probability of transitioning between different biological states over time. This mathematical framework is used to govern events like DNA damage accumulation, repair processes, and cell-cycle progression within the model.
Inverse Dose Rate Effect
This phenomenon occurs when cells are irradiated in multiple fractions with a long time gap between them. The model shows this effect arises from the interplay between sublethal damage repair and cell redistribution into radioresistant phases during the inter-fraction period, leading to non-monotonic survival probability.
Stochastic Variability
Because individual cell trajectories are simulated rather than just averaged, the model generates distributions of possible outcomes. This captures intra-cellular stochasticity in DNA damage and repair, showing how individual cells evolve differently following irradiation.

Terminology

Summary

Accurately linking microscopic energy deposition from ionizing radiation to emergent biological outcomes remains a central challenge in radiobiological modeling, particularly when stochastic damage induction, cell-cycle dynamics, and spatial organization within irradiated tissues must be treated explicitly and consistently across scales<ref:2604. p. 0

Where the paper says it

A stochastic agent-based extension of the GSM2 model for particle therapy: cell-cycle dynamics, dose-rate dependence, and fractionation effects Francesco G. Cordoni1, 2, 3, *, Marco Battestini1 , and Marta Missiaggia4 1Trento Institute for Fundamental Physics and Application (TIFPA), via Sommarive 15, Trento, 38123, Italy 2Department of Civil, Environmental and Mechanical Engineering, University of Trento, via Mesiano 77, Trento, 38123, Italy 3Department of Civil and Environmental Engineering, Louisiana State University (LSU), 4905 S Quad Dr, Baton Rouge, LA 70808, USA 4Department of Physics & Astronomy, Louisiana State University (LSU), 202 Nicholson Hall, Baton Rouge, LA 70803, USA July 10, 2026 Abstract Accurately linking microscopic energy deposition from ionizing radiation to emergent biological outcomes remains a central challenge in radiobiological modeling, particularly when stochastic damage induction, cell-cycle dynamics, and spatial organization within irradiated tissues must be treated explicitly and consistently across scales.

27630#pg2> The framework is applied to three-dimensional tumor spheroids irradiated with 1H and 12C ions across a range of energies and dose rates<ref:2604. p. 0
Where the paper says it

Radiation-induced damage induction, repair, misrepair, cell-cycle progression, proliferation, and migration are treated as competing stochastic events resolved using a next-event, event-driven algorithm that scales computationally with system size while preserving full single-cell resolution. The framework is applied to three-dimensional tumor spheroids irradiated with 1H and 12C ions across a range of energies and dose rates.

27630#pg10> Several empirically established trends in biological response, including the dose-rate dependence of cell survival, its attenuation at high LET, and the inverse dose rate effect in split-dose irradiation, emerge from the model through the explicit coupling of particle arrivals, damage accumulation, and repair kinetics<ref:2604. p. 0
Where the paper says it

We characterize the spatiotemporal evolution of cell-cycle phase composition and spheroid volume following irradiation, and examine the dependence of cell survival on dose rate over four orders of magnitude, from 10−5 Gy s−1 to 10−2 Gy s−1 . Several empirically established trends in biological response, including the dose-rate dependence of cell survival, its attenuation at high LET, and the inverse dose rate effect in split-dose irradiation, emerge from the model through the explicit coupling of particle arrivals, damage accumulation, and repair kinetics, without recourse to empirical correction factors as typically done.

27630#pg10> The model captures repair, reassortment, and repopulation within a single consistent framework and provides a basis for quantifying stochastic variability in treatment response at the single-cell level<ref:2604. p. 0
Where the paper says it

Several empirically established trends in biological response, including the dose-rate dependence of cell survival, its attenuation at high LET, and the inverse dose rate effect in split-dose irradiation, emerge from the model through the explicit coupling of particle arrivals, damage accumulation, and repair kinetics, without recourse to empirical correction factors as typically done. The model captures repair, reassortment, and repopulation within a single consistent framework and provides a basis for quantifying stochastic variability in treatment response at the single-cell level.

27630#pg10>.

Modeling Heterogeneity through Agent-Based Dynamics

The researchers developed an agent-based modeling framework that extends the GSM2 to explicitly resolve individual cells and their biological dynamics following particle irradiation<ref:2604.27630#pg9> Each cell is represented as an autonomous agent whose internal state evolves according to a continuous-time Markov chain (CTMC)<ref:2604. p. 3

Where the paper says it

2 Materials and Methods We developed an agent-based modeling framework that extends GSM2 to explicitly resolve individual cells and their biological dynamics following particle irradiation. Each cell is represented as an autonomous agent whose internal state evolves according to a continuous-time Markov chain (CTMC).

27630#pg9> System evolution is simulated using a next-event, event-driven algorithm, enabling efficient computation at single-cell resolution over large spatial domains. p. 3
Where the paper says it

Radiation-induced DNA damage, repair, misrepair, cell-cycle progression, proliferation, cell death, and active migration are treated as stochastic events governed by transition rates derived from GSM2 and biologically motivated cell-cycle kinetics. System evolution is simulated using a next-event, event-driven algorithm, enabling efficient computation at single-cell resolution over large spatial domains.

Incorporating Physical and Biological Scales

The framework integrates several physical and biological components to capture the complexity of radiation response<ref:2604.27630#pg9> The model explicitly includes:

  1. Radiation field modeling, where particle arrivals are sampled stochastically from an exponential distribution with a rate proportional to the prescribed dose rate. p. 3

    Where the paper says it

    The radiation field is characterized by the particle type, energy, LET, and temporal structure, with the beam propagating along the z-axis. The particle fluence is defined as F = D[Gy] 1:602 × 10−9 [Gy · cm2 (keV=—m)−1 ] · LET[keV=—] : Particle arrivals are modeled as stochastic events, with inter-arrival times sampled from an exponential distribution with rate proportional to the prescribed dose rate.

  2. Oxygenation, modeled as the quasi-steady-state solution of a reaction–diffusion equation to capture spatial heterogeneity. p. 4

    Where the paper says it

    tk ∼ Exp` d˙). 3 Oxygenation The oxygen distribution inside the spheroid is modeled as the quasi-steady-state solution of a reaction– diffusion equation.

  3. DNA damage induction, where sublethal lesions X and lethal lesions Y are induced based on LET and oxygenation through the Oxygen Enhancement Ratio (OER). p. 5

    Where the paper says it

    Damage induction is modulated by the oxygen fixation effect through the Oxygen Enhancement Ratio (OER): »(LET; [O2]) = k(LET) OER(LET; [O2]); with OER(LET; [O2]) = KO2 · KLETM + LET‚ KLET + LET‚ + [O2] KO2 + [O2] ; where M, KO2 , KLET, and ‚ are fitted parameters [Scifoni and Durante, 2013, Epp and Linard, 1972]. 6 Single-cell response model Given the spatial distribution of lethal (Y ) and sublethal (X) lesions within each domain d of cell j, damage evolution is governed by the following GSM2 reaction pathways [Cordoni et al.

Resolution of Stochasticity and Dose-Rate Effects

The event-driven simulation algorithm handles four competing event categories: repair and lethal lesion formation events, cell-cycle transitions and division, irradiation particle arrivals, and cell migration. This formulation allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics. p. 3

Where the paper says it

The temporal arrival of individual particles is sampled stochastically from the prescribed dose-rate distribution and coupled to biological repair processes through the shared event queue. This formulation is particularly suited to the study of dose-rate and fractionation effects, and allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics, without any ad hoc modification of the LQ formalism or use of empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942].

Mechanistic Validation and Clinical Relevance

The results demonstrate that carbon ions produce more severe cell killing than protons at the same energy per nucleon, consistent with their higher LET and biological effectiveness The split-dose simulations reveal a non-monotonic dependence of survival probability on inter-fraction time, which is mechanistically consistent with the inverse dose rate effect. Furthermore, the framework reproduces three of the classical Rs of radiobiology—repair, reassortment, and repopulation—within a single consistent framework. p. 3

Where the paper says it

We demonstrate its capabilities using three-dimensional tumor spheroids irradiated with protons and carbon ions across a range of energies and dose rates, analyzing spatiotemporal population dynamics, dose-rate dependence of cell survival, and split-dose fractionation effects. These results illustrate how the proposed framework provides a mechanistic, cell-population radiobiological model that accounts for several classical radiobiological phenomena, such as repair, reassortment, repopulation, and the inverse dose-rate effect, within a single, consistent framework applicable to both conventional and particle radiotherapy.

Limitations and Future Directions

Some limitations noted include cell fate being binary, omitting death modalities such as apoptosis and radiation-induced senescence Oxygenation is modelled through a quasi-static radial profile without dynamic coupling to vascular supply or nutrient transport. Future work will investigate the response of highly heterogeneous tissues, with particular attention to skin, whose acute and late toxicity is among the most clinically relevant normal-tissue complications in particle therapy.

Calibration of Biological Parameters

The GSM2 rate parameters (r; a; b) for each cell-cycle phase are calibrated by fitting the GSM2 survival curve to the phase-specific LQ response characterized by the (¸; ˛) values reported in Section 2.10 The agreement between the GSM2 and LQ curves is good across the dose range considered for all three cell-cycle phases, indicating that the calibrated GSM2 parameters consistently reproduce the dominant radiobiological features encoded in the LQ fits. p. 7

Where the paper says it

10 Model parameters The GSM2 model is parameterized by three rate constants per cell-cycle phase: the repair rate r, the conversion rate a (sublethal to lethal), and the binary misrepair rate b (pairwise interaction of sublethal lesions). These parameters are assigned independently for each phase of the cell cycle and are calibrated so that the resulting GSM2 survival curves reproduce the phase-specific linear-quadratic parameters (¸; ˛) reported in the experimental literature [Hall et al.

Dose-Rate Dependence Analysis

The quadratic parameter ˛ exhibits a pronounced dose-rate dependence, decreasing monotonically with decreasing dose rate and approaching zero at the lowest rates This behavior reflects the interplay between the rate of dose delivery, sublethal damage repair, and cell-cycle redistribution captured by the agent-based model.

Fractionation Effects

The survival probability exhibits a non-monotonic dependence on inter-fraction time for a 1:5 Gy + 1:5 Gy fractionation scheme, showing an initial increase due to sublethal damage repair, followed by a second increase associated with redistribution into the radioresistant S phase. This oscillatory behavior is consistent with the inverse dose rate effect reported experimentally for low-LET radiation. p. 0

Where the paper says it

Several empirically established trends in biological response, including the dose-rate dependence of cell survival, its attenuation at high LET, and the inverse dose rate effect in split-dose irradiation, emerge from the model through the explicit coupling of particle arrivals, damage accumulation, and repair kinetics, without recourse to empirical correction factors as typically done. The model captures repair, reassortment, and repopulation within a single consistent framework and provides a basis for quantifying stochastic variability in treatment response at the single-cell level.

Conclusion on Framework Utility

The framework provides a mechanistic, cell-population radiobiological model that accounts for several classical radiobiological phenomena, such as repair, reassortment, repopulation, and the inverse dose-rate effect<ref:2604.27630#pg10> The resulting dual stochasticity allows the model to generate distributions of possible outcomes rather than single deterministic predictions. p. 8

Where the paper says it

11 Model outputs and endpoints Model outputs include detailed single-cell fate histories, spatial distributions of DNA damage and survival, population growth curves, and survival probabilities. Clinically relevant endpoints such as TCP and NTCP, as well as treatment-response heterogeneity, can be derived by aggregating singlecell outcomes into population-level metrics, providing a mechanistic link between microscopic damage processes and macroscopic clinical predictions [Battestini et al.

Computational Efficiency

The computational scaling of the framework is demonstrated in Figure 3, where computation time grows approximately linearly with total cell number for a fixed dose of 1 Gy delivered by 1H ions at 100 MeV The dominant bottleneck is the irradiation step, whose cost scales with the number of simulated particles and therefore with the prescribed dose.

Modeling Cell Cycle Dynamics

The explicit representation of the cell cycle is a key feature of the framework, motivated by the well-established dependence of radiosensitivity on cell-cycle phase Each cell progresses through the cycle independently, with phase durations sampled from Gamma distributions and radiosensitivity parameters assigned according to the current phase.

Spatial Organization

The radial stratification of cell-cycle phases, with a quiescent core surrounded by a cycling shell, is resolved at the single-cell level and evolves dynamically following irradiation This spatial organization reflects the interplay between radiation-induced damage, repair kinetics, and cell-cycle progression.

General Applicability

By embedding GSM2 within a next-event, continuous-time Markov architecture, the model resolves individual cells across all stages of the radiation response—from local energy release and DNA damage induction through repair, cell-cycle progression, and spatial reorganization. The biological accuracy of the model is inherited from GSM2, whose predictions for cell survival have been validated against experimental data for multiple cell lines and irradiation conditions. p. 1

Where the paper says it

In conventional photon radiotherapy, the linear-quadratic (LQ) model remains the dominant framework for describing cell survival and guiding clinical dose prescription and fractionation [McMahon, 2018, Fowler, 1989]. While the LQ model is well supported empirically and computationally convenient, it is phenomenological in nature: its parameters ¸ and ˛ are fitted to experimental survival data and do not derive from an explicit description of the underlying damage and repair processes.

Stochastic Variability

The spread of individual trajectories around the ensemble mean is largest immediately after irradiation and narrows over time as repair proceeds This dual stochasticity allows the model to generate distributions of possible outcomes rather than single deterministic predictions, which is relevant for quantifying uncertainty in treatment response under realistic irradiation conditions.

Model Parameters and Rates

The GSM2 model is parameterized by three rate constants per cell-cycle phase: the repair rate r, the conversion rate a (sublethal to lethal), and the binary misrepair rate b (pairwise interaction of sublethal lesions) These parameters are assigned independently for each phase of the cell cycle and are calibrated so that the resulting GSM2 survival curves reproduce the phase-specific linear-quadratic parameters (¸; ˛) reported in the experimental literature.

Model Outputs

Model outputs include detailed single-cell fate histories, spatial distributions of DNA damage and survival, population growth curves, and survival probabilities Clinically relevant endpoints such as TCP and NTCP can be derived by aggregating single-cell outcomes into population-level metrics.

Simulation Environment

All simulations were performed on a workstation equipped with an Intel® CoreTM i9-13900K processor (32 logical cores, 2 threads per core, maximum clock speed 5:8 GHz) and 62 GiB of RAM. The framework is implemented in Julia, version 1. p. 8

Where the paper says it

0-19-generic, x86_64). The framework is implemented in Julia, version 1.

12.1.

Reference to Specific Results

The comparison with the G1-only and no-repopulation reference cases in Figure 14 helps to separate the contributions of individual mechanisms The second survival increase is absent in the G1-only case, confirming its association with redistribution into S phase.

The characteristic oscillatory structure of the survival curve, including an initial increase due to sublethal damage repair, a second increase associated with redistribution into the radioresistant S phase, and a subsequent decrease as cells re-enter the radiosensitive G2/M phases, is directly supported by the simulated cell-cycle distributions at the time of the second fraction.

The ¸ parameter of the linearquadratic model is found to be independent of dose rate across all ion species and energies considered, while ˛ decreases monotonically with decreasing dose rate, approaching zero at the lowest values. This behavior is substantially attenuated for high-LET carbon ions, where single-track lethal events dominate, and the quadratic contribution is already small at clinical dose rates. p. 9

Where the paper says it

Across the three panels, the mean lethal lesion count increases from protons at 80 MeV (Figure 9a) to protons at 30 MeV (Figure 9b) to carbon ions at 80 MeV=u (Figure 9c), consistent with the increasing LET of the respective ion species and kinetic energy. Figure 10 shows the survival probability as a function of delivered dose for three ion species and energy combinations, each resolved across a range of dose rates spanning four orders of magnitude, from 10−5 Gy s−1 to 10−2 Gy s−1 , where the upper limit corresponds to the clinical dose rate range.

The split-dose results of Figures 13 and 14 illustrate the inter-fraction time dependence of survival probability for a 1:5 Gy + 1:5 Gy fractionation scheme The characteristic oscillatory structure of the survival curve, including an initial increase due to sublethal damage repair, a second increase associated with redistribution into the radioresistant S phase, and a subsequent decrease as cells re-enter the radiosensitive G2/M phases, is directly supported by the simulated cell-cycle distributions at the time of the second fraction.

The oscillatory structure of the phase curves confirms the mechanistic origin of survival trends described above. The comparison with reduced model configurations isolates the distinct contributions of repair, reassortment, and repopulation, three of the classical Rs of radiobiology. p. 3

Where the paper says it

We demonstrate its capabilities using three-dimensional tumor spheroids irradiated with protons and carbon ions across a range of energies and dose rates, analyzing spatiotemporal population dynamics, dose-rate dependence of cell survival, and split-dose fractionation effects. These results illustrate how the proposed framework provides a mechanistic, cell-population radiobiological model that accounts for several classical radiobiological phenomena, such as repair, reassortment, repopulation, and the inverse dose-rate effect, within a single, consistent framework applicable to both conventional and particle radiotherapy.

The full ABM diverges from this reference at intermediate and late inter-fraction times, a difference that would not be captured by population-level approaches. Taken together, these results show that the model reproduces three of the classical Rs of radiobiology [Hall et al. p. 3

Where the paper says it

We demonstrate its capabilities using three-dimensional tumor spheroids irradiated with protons and carbon ions across a range of energies and dose rates, analyzing spatiotemporal population dynamics, dose-rate dependence of cell survival, and split-dose fractionation effects. These results illustrate how the proposed framework provides a mechanistic, cell-population radiobiological model that accounts for several classical radiobiological phenomena, such as repair, reassortment, repopulation, and the inverse dose-rate effect, within a single, consistent framework applicable to both conventional and particle radiotherapy.

, 2006, Inaniwa et al. p. 1
Where the paper says it

In particle therapy, the situation is further complicated by the strong dependence of biological effectiveness on particle type, energy, and LET, which the LQ model cannot account for without modification. RBE-weighted dose models, such as the Microdosimetric Kinetic Model (MKM) and its variants, are widely used in clinical practice to estimate cell survival and prescribe biologically equivalent doses [Inaniwa and Kanematsu, 2018, Bellinzona et al.

, 2013].

The presence of phase-dependent heterogeneity has been identified as a relevant contributor to inter-tumour variability in treatment response [Inaniwa et al., 2024]. The resulting phase-dependent differential killing is illustrated in Figures 5, 6, and 7 without requiring any empirical adjustment. p. 3

Where the paper says it

This formulation is particularly suited to the study of dose-rate and fractionation effects, and allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics, without any ad hoc modification of the LQ formalism or use of empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942]. The resulting framework enables efficient simulation of large cell populations under particle irradiation with full single-cell resolution, capturing both intra-cellular stochasticity in DNA damage and repair and inter-cellular heterogeneity arising from microdosimetric variability, cell-cycle state, and local microenvironmental conditions.

The model reproduces the dose-rate dependence of cell survival without empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942]. The resulting dose-rate effects are shown in Figs. p. 3

Where the paper says it

This formulation is particularly suited to the study of dose-rate and fractionation effects, and allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics, without any ad hoc modification of the LQ formalism or use of empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942]. The resulting framework enables efficient simulation of large cell populations under particle irradiation with full single-cell resolution, capturing both intra-cellular stochasticity in DNA damage and repair and inter-cellular heterogeneity arising from microdosimetric variability, cell-cycle state, and local microenvironmental conditions.

10 and 11. The Γ parameter of the LQ model is found to be independent of dose rate across all ion species and energies considered, while the α parameter decreases monotonically with decreasing dose rate, approaching zero at the lowest values. p. 9
Where the paper says it

Across the three panels, the mean lethal lesion count increases from protons at 80 MeV (Figure 9a) to protons at 30 MeV (Figure 9b) to carbon ions at 80 MeV=u (Figure 9c), consistent with the increasing LET of the respective ion species and kinetic energy. Figure 10 shows the survival probability as a function of delivered dose for three ion species and energy combinations, each resolved across a range of dose rates spanning four orders of magnitude, from 10−5 Gy s−1 to 10−2 Gy s−1 , where the upper limit corresponds to the clinical dose rate range.

This is consistent with the standard interpretation of α as arising from pairwise interactions between sublethal lesions from different tracks [McMahon, 2018].

The dose-rate dependence of survival is substantially attenuated for high-LET carbon ions, particularly at low energies, where single-track lethal events dominate. These trends emerge from the model without any ion-specific tuning of the dose-rate response, which serves as a useful consistency check on the underlying formulation. p. 3

Where the paper says it

The temporal arrival of individual particles is sampled stochastically from the prescribed dose-rate distribution and coupled to biological repair processes through the shared event queue. This formulation is particularly suited to the study of dose-rate and fractionation effects, and allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics, without any ad hoc modification of the LQ formalism or use of empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942].

The interaction between protracted irradiation and cell-cycle dynamics is illustrated in Figure 12 At low dose rates, the irradiation window spans multiple cell-cycle durations, producing a characteristic phase arrest visible in Figure 12b This coupling between dose rate and cell-cycle kinetics is not captured by population-level models and provides additional insight into the biological mechanisms underlying dose-rate effects.

The inclusion of oxygenation is a further source of spatial heterogeneity in radiosensitivity that is included in the current architecture The oxygen distribution is modelled as a quasi-static, radially symmetric profile and modulates lesion induction rates via the OER (Section 2.3).

Any radial oxygen profile can be prescribed, although dynamic reoxygenation is not yet implemented and is left as a direction for future development. A detailed investigation of the oxygen effect is deferred to [Battestini et al. p. 8

Where the paper says it

11 Model outputs and endpoints Model outputs include detailed single-cell fate histories, spatial distributions of DNA damage and survival, population growth curves, and survival probabilities. Clinically relevant endpoints such as TCP and NTCP, as well as treatment-response heterogeneity, can be derived by aggregating singlecell outcomes into population-level metrics, providing a mechanistic link between microscopic damage processes and macroscopic clinical predictions [Battestini et al.

, 2026].

The model parameters are drawn from literature values rather than fitted to experimental spheroid data, and a systematic quantitative comparison with experimental measurements remains to be carried out.

The inclusion of biochemical pathways governing permanent cell-cycle arrest, in particular the p53/p21 axis, and of dynamic oxygen transport coupled to an explicit vascular description are the primary planned developments. The model architecture is also compatible with the direct computation of tumor control probability and normal tissue complication probability by aggregating single-cell outcomes, providing a route towards mechanistically grounded predictions of clinically relevant treatment endpoints. p. 8

Where the paper says it

11 Model outputs and endpoints Model outputs include detailed single-cell fate histories, spatial distributions of DNA damage and survival, population growth curves, and survival probabilities. Clinically relevant endpoints such as TCP and NTCP, as well as treatment-response heterogeneity, can be derived by aggregating singlecell outcomes into population-level metrics, providing a mechanistic link between microscopic damage processes and macroscopic clinical predictions [Battestini et al.

The framework was developed as an explicit single-cell extension of the Generalized Stochastic Microdosimetric Model (GSM2)<ref:2604.27630#pg9> Each cell is represented as an autonomous agent whose internal state evolves according to a continuous-time Markov chain driven by GSM2 transition rates. p. 0

Where the paper says it

To address this, we introduce a stochastic agent-based radiobiological modeling framework for simulating biological response to particle irradiation, developed as an explicit single-cell extension of the Generalized Stochastic Microdosimetric Model (GSM2 ). Each cell is represented as an autonomous agent whose internal state, including DNA lesion counts, cellcycle phase, and oxygenation level, evolves according to a continuous-time Markov chain driven by GSM2 transition rates.

The next-event algorithm identifies the earliest scheduled event across all agents and irradiation events, advances the global clock to that time, updates the system state, and reschedules only the affected events. This process ensures that exact continuous-time dynamics are preserved without any fixed time step. p. 7

Where the paper says it

The global clock advances as t ← t + fi, and the corresponding event is executed. Only the event times of cells whose neighborhood changes are resampled; all others are decremented by fi, preserving exact continuous-time dynamics without any fixed time step.

The framework is implemented in Julia, version 1.12.1 The source code is publicly available at the GitHub repository: https://github. p. 8

Where the paper says it

1 [Bezanson et al. The source code is publicly available at the GitHub repository: https://github.

com/francescogcordoni/GSM2 Julia. p. 8
Where the paper says it

The source code is publicly available at the GitHub repository: https://github. com/francescogcordoni/GSM2_Julia.

The computational scaling of the framework is demonstrated in Figure 3, where computation time grows approximately linearly with total cell number for a fixed dose of 1 Gy delivered by 1H ions at 100 MeV The dominant bottleneck is the irradiation step, whose cost scales with the number of simulated particles and therefore with the prescribed dose.

The model captures repair, reassortment, and repopulation within a single consistent framework and provides a basis for quantifying stochastic variability in treatment response at the single-cell level<ref:2604.27630#pg10>. The resulting framework enables efficient simulation of large cell populations under particle irradiation with full single-cell resolution, capturing both intra-cellular stochasticity in DNA damage and repair and inter-cellular heterogeneity arising from microdosimetric variability, cell-cycle state, and local microenvironmental conditions<ref:2604. p. 3

Where the paper says it

This formulation is particularly suited to the study of dose-rate and fractionation effects, and allows well-known trends in cell survival as a function of dose rate, including the attenuation of the quadratic component of the LQ response and the inverse dose rate effect, to emerge directly from the interplay between particle arrivals, damage accumulation, and repair kinetics, without any ad hoc modification of the LQ formalism or use of empirical correction factors such as the Lea–Catcheside factor [Lea and Catcheside, 1942]. The resulting framework enables efficient simulation of large cell populations under particle irradiation with full single-cell resolution, capturing both intra-cellular stochasticity in DNA damage and repair and inter-cellular heterogeneity arising from microdosimetric variability, cell-cycle state, and local microenvironmental conditions.

27630#pg10>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Explicit single-cell resolution allows for capturing intra-cellular stochasticity in DNA damage and repair, which is crucial for modeling how individual cells evolve through distinct biological states following irradiation. This enables the simulation to capture cell-to-cell variability in damage evolution as illustrated by the spread of individual trajectories around the ensemble mean.

  2. Integration of cell-cycle progression and spatial organization into an agent-based framework allows the model to simulate spatiotemporal population dynamics, dose-rate dependence of cell survival, and split-dose fractionation effects. This provides a mechanistic link between microscopic stochasticity in energy deposition propagates through biological repair pathways and manifests as macroscopic variability in tumor control and normal tissue toxicity.

  3. The model can predict clinically relevant endpoints such as tumour control probability (TCP) and normal tissue complication probability (NTCP) by aggregating single-cell outcomes into population-level metrics, providing a mechanistic link between microscopic damage processes and macroscopic clinical predictions.

Abstract

We present a stochastic agent-based extension of the Generalized Stochastic Microdosimetric Model (GSM squared) that links microscopic energy deposition by ionizing radiation to population-scale response. Each cell is represented as an agent whose DNA damage, cell-cycle phase, and fate evolve as a continuous-time Markov process. Particle arrivals, repair, misrepair, division, death, and migration are treated as competing stochastic events in an event-driven algorithm, enabling track-structure simulations of spheroids with up to 1.5 times10 5 cells in minutes on a single workstation. For static populations, the model reproduces the Lea--Catcheside dose-rate dependence of survival for proton and carbon-ion irradiation over four orders of magnitude in dose rate. Dose-rate sparing decreases with increasing LET and disappears for low-energy carbon ions. After calibration to a single survival curve for aerobic X-ray irradiation of V79 spheroids, the model predicts without further fitting survival under anoxia and after carbon-ion irradiation in aerobic and anoxic conditions, reproducing measured oxygen enhancement ratios and relative biological effectiveness. Applied to three-dimensional tumor spheroids, the framework resolves responses to single, split, and fractionated irradiations, separates the effects of repair, reassortment, and repopulation, and quantifies stochastic variability between realizations. The model provides a mechanistic single-cell framework for studying time-dependent effects and uncertainty in particle irradiation.

Sources

Related papers