Establishing massively parallel models to examine the influence of cell heterogeneity on tumor growth

Parallel 3D cellular automaton models of tumor growth can efficiently capture emergent morphology. We extended a 2D growth model to 3D to examine the influence of symmetric division in heterogeneous tumors on growth dynamics. As extending to 3D severely increased time-to-solution, we parallelized the model using N-body, lattice halo exchange, and adaptive communication schemes. Supplementing prior work from Tanade et al. (2022), we demonstrated over 55x speedup and evaluated performance on ≤30 nodes of Stampede2. This work established a framework to parametrically study 3D growth dynamics, and of the cancer phenotypes we studied, the parallel model better scaled when tumor boundaries were radially symmetric.

Duke Scholars

Author Amanda Randles Biomedical Engineering