Continuous and Discrete Models of Melanoma Progression Simulated in Multi-GPU Environment

Existing computational models of cancer evolution mostly represent very general approaches for studying tumor dynamics in a homogeneous tissue. Here we present two very different cancer models: the heterogeneous continuous/discrete and purely discrete one, focusing on a specific cancer type – melanoma. This tumor proliferates in a complicated heterogeneous environment of the human skin. The results from simulations obtained for the two models are confronted in the context of their possible integration into a single multi-scale system. We demonstrate that the interaction between the tissue – represented by both the concentration fields (the continuous model) and the particles (the discrete model) – and the discrete network of blood vessels is the crucial component, which can increase the simulation time even one order of magnitude. To compensate this time lag, we developed GPU/CUDA implementations of the two melanoma models. Herein, we demonstrate that the continuous/discrete model, run on a multi-GPU cluster, almost fifteen times outperforms its multi-threaded CPU implementation.