Each year, cancer is responsible for 13% of all deaths worldwide. In the United States, that percentage increases to 25%. Given these statistics, there has been growing interest in developing computer simulations capable of modeling cancer and the way in which it spreads. It is hoped that someday these simulations will allow for greater understanding of complex cellular biology as well as provide an efficient, low-cost method for evaluating the efficacy of various treatments.
Description
Current cancer simulations consist of force-based models. Policies for these models are derived from a complex set of rules and equations that must be generated by hand. Our goal is to leverage multi-agent techniques to create a model that is more robust and flexible. Our domain features both healthy cells and cancerous cells embodied as agents whose sequential decision making is modeled as a Markov decision process (MDP). By constructing a reward function, we can establish the intentions of our healthy cells. We then use value iteration, a backward induction algorithm, to automatically generate the optimal policy for these cells. Manipulating the reward function allows us to generate a different policy for the cancerous cells. Finally, these policies are fed into our simulation tool which allows for the visualization of the interaction between the various cells.
Publication(s)
Matthew Brown, Emma Bowring, Shira Epstein, Mufaddal Jhaveri, Rajiv Maheswaran, Parag Mallick, Shannon Mumenthaler, Michelle Povinelli, and Milind Tambe. Applying Multi-Agent Techniques to Cancer Modeling, In Proceedings of the Sixth Workshop on Multiagent Sequential Decision Making in Uncertain Domains (at AAMAS-11), May 2011.[pdf]
