Development of Gell: A GPU-Powered Ultra-Large-Scale Cell-Based Simulator
J Du1*, Y Zhou2, L Jin2, K Sheng1, (1) UCLA School of Medicine, Los Angeles, CA, (2) UCLA School of Engineering, Los Angeles, CA
Presentations
PO-GePV-M-10 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: To understand macroscopic imaging feature formation as a result of underlying cancer cell biology and growth dynamics, we build a novel fast GPU-powered cell-based simulation platform-Gell suited for ultra-large-scale tumor growth modeling.
Methods: Tumor development is a multifaced and multiscale problem. At the biochemical microenvironment level, we implemented a fast paralleled PDE solver for the secretion, diffusion, uptake, and decay of oxygen and Vascular Endothelial Growth Factor (VEGF) with computational complexity proportional to volume voxel number. At the single-cell level, we included modules to describe cell proliferation, growth, death, nutrient uptake, and angiogenic factor release. At the multicellular level, we described the mechanical interactions for cell-cell adhesive and repulsion with computational complexity proportional to the number of cells. We combined these modules to simulate HCC827 non-small cell lung carcinoma spheroids' growth and compared the results with state-of-the-art CPU simulator PhysiCell.
Results: Gell simulation results of the approximately constant spheroid diameter growth speed and the crack-like necrotic core microstructure are consistent with PhysiCell simulation and in vitro experimental observation. However, the computational time for 18 days of in vitro spheroid growth was reduced from 76 hours using CPU-based PhysiCell (dual 6-core Intel Xeon X5690) to seven minutes using GPU-based Gell (Nvidia RTX 2080Ti). In a large multicellular simulation test for a 1-hour biological process of 16 million cells, the simulation was accelerated by over one thousand times compared with PhysiCell.
Conclusion: In this work, we developed a fast GPU-based simulator, Gell, for avascular multicellular processes. The GPU implementation afforded 1000X acceleration over the CPU method. The acceleration is essential to overcoming the bottleneck in CPU simulation for ultra-large-scale, morphologically complex centimeter-level tumors and tumor vasculature that form medical image features.
ePosters
Keywords
Taxonomy
IM/TH- Foundational Skills: Hardware acceleration and parallel processing
Contact Email