Trajectory Landscapes for Therapeutic Strategy Design in Agent-Based Tumor Microenvironment Models
New framework uses AI simulation to predict cancer progression and optimize treatment timing without longitudinal patient data.
A team from Oregon Health & Science University, led by Young Hwan Chang, has published a novel AI framework titled 'Trajectory Landscapes for Therapeutic Strategy Design in Agent-Based Tumor Microenvironment Models.' The research addresses a critical bottleneck in oncology: clinical datasets from multiplex tissue imaging (MTI) are typically single snapshots in time, making it impossible to directly observe how a tumor evolves or determine the optimal moment for intervention. The team's solution leverages detailed, mechanistic Agent-Based Models (ABMs)—stochastic simulators of cellular interactions in the tumor microenvironment. By running thousands of simulations with systematic parameter variations, they generate a comprehensive 'trajectory landscape' of possible cancer progression paths.
The core innovation is reducing this high-dimensional simulation data into an actionable model. Using time-series data from simulations, the team learns a probabilistic Markov State Model (MSM). This MSM identifies metastable states (like periods of tumor quiescence or aggression) and the probabilities of transitioning between them. A clinician can then map a real patient's single MTI snapshot onto this pre-computed landscape. The framework further conditions the MSM on key biological parameters to create patient-group-specific models, which are used to formulate a finite-horizon Markov Decision Process (MDP). This MDP mathematically defines the optimal treatment scheduling policy, effectively using AI simulation to ground therapeutic decisions in predicted long-term outcomes, all without needing longitudinal data from that specific patient.
- Uses Agent-Based Models to simulate thousands of possible tumor microenvironment trajectories and create a predictive 'landscape'.
- Learns a probabilistic Markov State Model from simulations to identify key cancer progression states and transition probabilities.
- Formulates treatment scheduling as a Markov Decision Process, enabling optimized, personalized therapy design from a single patient snapshot.
Why It Matters
Enables data-driven, personalized cancer treatment scheduling by predicting disease progression from a single biopsy, potentially improving outcomes and avoiding ineffective therapies.