Are Tumors Intelligent? - A learning-based approach to tumor invasion

Tumor spheroids embedded in fiber networks are an important in-vitro experimental setup for studying cancer systems. We develop a three-dimensional computational model to capture the dynamics of such embedded systems. This model consists of a 3D vertex model spheroid, coupled to an elastic fiber network via actively contracting linear springs. We quantify the coupled mechanical behavior of such systems, in particular the resulting dynamical remodeling of the fiber network, which is an observed feature in experimental systems. We further investigate how the effectiveness in remodeling the fiber network may be enhanced by introducing physical learning algorithms to evaluate efficient coupling prescriptions. Interestingly, the learning dynamics within this system bear resemblance to an aging process; the fiber networks undergo relaxation in response to the continuous application of feedback forces by the coupled springs, especially when subjected to an input force. This gradual adaptation effectively encodes a memory of the input-output relationship, further enriching our understanding of these complex systems. The introduction of such physical learning rules is a step towards modeling tumor spheroids as active collectives that efficiently align their extracellular matrices to induce invasion.