Poster

Machine Learning based Parametrization of Large Scale Tumor Simulations

Karlsruhe Institute of Technology, Steinbuch Centre of Computing, Junior Research Group Multiscale Biomolecular, Hermann-von-Helmholtz-Platz 1, 76344, Eggenstein-Leopoldshafen

Contact:  | Website

Machine Learning based parametrization of tumor simulations

Despite decades of substantial research, cancer remains a ubiquitous
scourge in the industrialized world. Effective treatments require a
thorough understanding of macroscopic cancerous tumor growth out
of individual cells in the tissue and microenvironment context.
Here, we aim to introduce the critical scale-bridging link between
clinical imaging and quantitative experiments focusing on small clus-
ters of cancerous cells by applying machine learning to drive model
building between them. We deploy Cells in Silico (CiS), a high per-
formance framework for large-scale tissue modeling developed by us.
Based on both a cellular potts model and an agent-based layer, CiS is
capable of accurately representing many physical and biological prop-
erties, such as individual cell shapes, cell division, cell motility etc.
The strong representational capacity of our model comes with the
need to adjust a large number of parameters according to experimental
findings. We present a generalized approach to optimize these param-
eters which allows the use of different sources of experimental data.
One major hurdle to achieve this goal is finding appropriate objective
functions. To overcome this we implemented a variation of the Parti-
cle Swarm Optimization algorithm which learns the objective function
during the optimization process.