Related Experiment Video
Updated: Aug 6, 2026
08:41
Modeling Chemotherapy Resistant Leukemia In Vitro
Published on: February 9, 2016
Reinforcement learning for chemotherapy scheduling in a stochastic tumor evolution model
1University of Southern California, Department of Aerospace & Mechanical Engineering, Los Angeles, California 90089-1191, USA.
Physical Review. E
|July 24, 2026
Summary
We developed a Q-learning framework to optimize chemotherapy schedules, balancing drug pressure and resistance evolution. This adaptive control strategy aims to delay tumor resistance by promoting subpopulation coexistence.
Area of Science:
- Computational Biology
- Cancer Research
- Evolutionary Dynamics
Background:
- Tumor evolution under chemotherapy drives drug resistance.
- Chemotherapy dosing schedules significantly impact treatment efficacy.
- Understanding tumor cell competition and resistance mechanisms is crucial.
Purpose of the Study:
- To optimize chemotherapy dosing schedules using Q-learning.
- To develop adaptive control strategies for managing tumor evolution.
- To delay the emergence and fixation of drug resistance.
Main Methods:
- Q-learning framework applied to a stochastic finite-cell tumor model.
- Markov process modeling tumor subpopulations (chemosensitive and resistant).
- Reward function engineered to promote coexistence and penalize imbalance.
Main Results:
- Optimal dosing policies derived to balance therapeutic pressure and resistance.
- Analysis of policies revealed dominant evolutionary trajectories.
- Robustness to partial observability and simplified heuristics were constructed.
Conclusions:
- Model-free adaptive control can steer tumor evolution away from resistance.
- Q-learning offers a promising approach for personalized cancer therapy.
- Balancing treatment and evolutionary dynamics is key to long-term control.
