Programmable models of growth and mutation of cancer-cell populations

Luca Bortolussi, Alberto Policriti

TL;DR

This paper introduces a novel programming language, sCCP, designed for building programmable, stochastic, and hybrid models of cancer cell populations. It demonstrates how this framework can be used to systematically reconstruct and analyze complex tumor growth models, offering a more flexible and detailed approach than traditional differential equation models, particularly for understanding the role of noise and external influences.

What Problem Does This Paper Address?

Modeling cancer cell population dynamics is crucial for understanding disease progression and developing effective therapies. Traditional approaches often rely on Ordinary Differential Equations (ODEs), which can be rigid and struggle to capture the inherent stochasticity and complex interactions (like drug delivery schedules) that characterize cancer development. The authors aim to bridge this gap by proposing a new modeling paradigm that is:

  • Programmable: Allows for high-level descriptions of models as networks of interacting agents.
  • Hybrid: Can incorporate both discrete (e.g., cell division, drug on/off states) and continuous (e.g., hormone concentration changes) dynamics.
  • Stochastic: Explicitly models random events and variations, which are common in biological systems and can influence outcomes.

Specifically, they focus on prostate cancer as a case study, aiming to model its growth, mutation, and response to therapies like intermittent androgen suppression, while also investigating the origins of noise observed in experimental data.

Methods

The core of the paper is the introduction and application of stochastic Concurrent Constraint Programming (sCCP). sCCP is a modeling language that combines agents and constraints, allowing for asynchronous communication and interaction through a shared "constraint store."

Key methodological contributions include:

  • Extending sCCP: The authors extend sCCP with primitives to handle:
  • Events: SBML-like events triggered by conditions on system state or simulation time, enabling the modeling of drug dispensation policies.
  • Random Updates: The ability to assign random values to variables, useful for modeling parameter uncertainty or intrinsic variability.
  • Hybrid Semantics: They define a hybrid semantics for sCCP by mapping sCCP programs to Transition-Driven Stochastic Hybrid Automata (TDSHA). This allows for a flexible interpretation of sCCP programs, ranging from fully discrete (CTMC) to fully continuous (ODE) models, with intermediate hybrid representations.
  • Agent-Based Modeling: The approach models populations as networks of interacting agents, where each agent represents a specific process (e.g., cell growth, hormone production). This contrasts with ODEs that typically track population-level quantities.
  • Prostate Cancer Model Construction: They demonstrate how to translate differential equations from existing prostate cancer models into sCCP agent networks, identifying variables and interactions.

Datasets

Not discussed in the paper. The paper focuses on developing a modeling framework and applying it to existing mathematical models of prostate cancer.

Key Insights and Ideas

  • Programmability for Biological Complexity: sCCP offers a powerful way to build complex biological models by composing simpler interacting agents, making models more modular and understandable than monolithic ODEs.
  • Noise Origin Investigation: The framework allows for distinguishing between noise arising from internal stochasticity (e.g., random cell division) versus external factors, which is crucial for interpreting experimental data.
  • Hybrid Nature of Cancer Dynamics: Cancer progression and treatment response often involve discrete events (e.g., drug on/off) and continuous processes (e.g., hormone levels). sCCP's hybrid semantics naturally captures this.
  • Low Population Dynamics: The stochastic, discrete nature of sCCP models allows for studying scenarios with small cell populations, where random extinction events or different behaviors can emerge, which is difficult with ODEs.
  • Limitations of Phenomenological Models for Noise: The paper suggests that simple phenomenological models, even when made stochastic, may not be rich enough to fully explain observed biological noise, implying a need for more detailed mechanistic models.

Key Results

  • Successful sCCP Implementation: The authors successfully developed sCCP programs that model prostate cancer growth under different drug dispensation policies (Continuous Androgen Suppression - CAS, and Intermittent Androgen Suppression - IAS).
  • Comparison with ODEs: Simulated trajectories from the sCCP models closely matched those from corresponding ODE models for large cell populations, indicating that the stochastic model accurately captures the average behavior.
  • Noise in Small Populations: For small cell populations, the stochastic sCCP model exhibited more pronounced noise and different behaviors (e.g., potential extinction of AI cells), which is not captured by ODEs.
  • External Noise Source Impact: Introducing an external noise source (as described in Figure 3(d)) led to observable noise even in large populations, highlighting the importance of considering external factors.
  • Limited Impact of Random PSA Production: Modifying the PSA production rate to be a random variable did not significantly alter the overall noisy behavior in large populations, suggesting that this specific source of randomness might be averaged out.

Limitations and Future Work

  • Hybrid Semantics for Complex sCCP: While the paper defines a hybrid semantics for a subset of sCCP, a fully satisfactory definition for more complex sCCP programs (e.g., with extensive parallel composition and local variable declarations) is still an open issue.
  • Detailed Biological Mechanisms: The paper acknowledges that phenomenological models may lack the richness to explain observed noise. Future work should involve incorporating more detailed biological mechanisms.
  • Spatial Organization: The current models do not explicitly account for the spatial organization of tumors. Future work will investigate this aspect.
  • Comparison with Experimental Data: While the paper mentions comparisons with real data for ODE models, further investigation and comparison of the sCCP models with experimental data are planned.
  • Advanced Analysis Techniques: The authors suggest that advanced techniques like statistical model checking could further enhance the analysis capabilities of the sCCP framework.

Summary updated

BibTeX
@inproceedings{bortolussi2011programmable,
  author = {Luca Bortolussi and Alberto Policriti},
  title = {{Programmable models of growth and mutation of cancer-cell populations}},
  booktitle = {CompMod},
  year = {2011},
  url = {https://openreview.net/forum?id=98JisDmTY3z},
}
Figure