Trajectory Landscapes for Therapeutic Strategy Design in Agent-Based Tumor Microenvironment Models.

Area of Science:

• Computational Biology
• Systems Biology
• Oncology

Background:

• Multiplex tissue imaging (MTI) provides spatial insights into the tumor microenvironment (TME).
• Clinical MTI data is often limited temporally, hindering understanding of TME spatiotemporal dynamics and optimal intervention timing.
• Agent-based models (ABMs) simulate TME evolution but face challenges in control design due to high dimensionality and parameter uncertainty.

Purpose of the Study:

• To develop a simulation-driven framework for designing therapeutic strategies using agent-based model (ABM) trajectory ensembles.
• To create a reduced-order model that captures TME evolution and links simulation dynamics to clinical observations.
• To enable adaptive, state-aware therapeutic strategies in oncology.

Main Methods:

• Generated an ensemble of simulated TME trajectories by perturbing ABM parameters.
• Constructed a low-dimensional trajectory landscape and identified a switched Markov State Model (MSM) from spatial summary statistics.
• Mapped patient MTI snapshots onto the landscape and formulated a Markov Decision Process (MDP) for intervention design.

Main Results:

• The framework successfully captures metastable states and transitions within the TME.
• MSM modes were linked to parameter regimes predictive of terminal-state outcomes.
• Patient data concordance with simulated phenotypes and outcomes was assessed.
• A finite-horizon MDP was formulated for reachability analysis and intervention design.

Conclusions:

• The developed framework enables simulation-grounded therapeutic policy design for partially observed biological systems.
• It allows for effective intervention design without requiring longitudinal patient measurements.
• This work represents a significant step towards adaptive, state-aware therapeutic strategies in oncology.