Adaptive Therapy for Metastatic Cancer: Mechanism vs Machine

Mathematical Biology

06 December 14:00 - 14:45

Alexander Anderson - Moffitt Cancer Center

Adaptive therapy is an evolution-based strategy that exploits heterogeneity via the cost of drug resistance. For resistant cells, benefits often exceed costs during therapy, however, in the absence of therapy, drug sensitive cells are fitter due to the cost of resistance. Treatment holidays exploit this cost by allowing sensitive cells to outcompete their resistant counterparts. A recent adaptive therapy clinical trial in metastatic prostate cancer, using abiraterone, has proven to be far more effective than continuous maximum tolerated dose (MTD). Here we will discuss new results in application to both metastatic melanoma and ovarian cancer. A central tenant of adaptive therapeutic approaches is that individual patient treatment response is the driver to stop or restart treatment. We have already developed evolutionary game theoretic, ordinary differential equation (ODE) and agent based models to predict optimal treatment timing and will discuss some of these here. Most recently we developed a state-of-the-art machine learning method that focuses solely on fine-grained prediction of tumor response, trained on in vitro experimental data from ovarian cancer spheroids. By using the evolving morphology of these spheroids, extracted from time-lapse microscopy with handcrafted computer vision techniques, we were able to characterize how the disease evolves in response to treatment. From this data we developed both an ODE model, that assumes both competition and phenotypic switching between the sensitive and resistant states, as well as a feed forward machine learning model to predict treatment response. We competed these two approaches in a long term adaptive therapy experiment and found that both are much more effective than the standard MTD. By uniting a mechanistic understanding of disease progression with data-driven machine learning models, it seems we can improve both patient outcomes and our understanding of the disease.
Mats Gyllenberg
University of Helsinki
Torbjörn Lundh
Chalmers/University of Gothenburg
Philip Maini
University of Oxford
Roeland Merks
Universiteit Leiden
Mathisca de Gunst
Vrije Universiteit Amsterdam


Roeland Merks


For practical matters at the Institute, send an e-mail to