Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

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


PO-GePV-M-21 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Cell-based simulation is powerful in studying tumor growth dynamics and understanding the causal relationship between microscopic cell behavior and macroscopic tissue textures. However, the high computational cost has hindered the development of large-scale cell-based models that is observable via medical imaging. To address this, we build a novel fast GPU-powered cell-based simulation platform-Gell suited for ultra-large-scale tumor growth modeling.

Methods: Our simulation platform uses physical principles to capture the complex spatiotemporal and multiscale biological and physical processes involved in tumor development. At the biochemical microenvironment level, we solved the partial differential equations (PDE) by a fast parallel Locally One-Dimensional (LOD) method to handle the spatial-temporal variation of diffusing substrates. We then adopted a faithful cell cycle model at the single-cell level that efficiently simulates cell proliferation, growth, and death. We developed a novel fast and memory-efficient Cell-Sorting Method for the most computationally intensive cell-cell mechanical interaction calculation for multicellular simulation. We combined the three modules to simulate hanging drop spheroid (HDS) growth up to one million cells and compared the results with the state-of-the-art CPU simulator PhysiCell.

Results: Gell accurately reproduced the nearly constant spheroid diameter growth speed and the crack-like necrotic core microstructure as shown in the PhysiCell simulation results and HDS in vitro models. In terms of computational performance, Gell simulation took 18 minutes vs. 43 hours using 16-thread CPU-based PhysiCell on a personal computer, translating into two orders of magnitude acceleration and one-tenth of the memory footprint.

Conclusion: In this work, we developed a fast GPU-based platform, Gell, for avascular multicellular tumor growth simulation. Gell affords ~100X acceleration over the CPU method with a fraction of the memory requirement. The platform paves the path for subsequent centimeter ultra-large-scale, morphologically complex, and vascularized tumor modeling that directly interprets tumor biology based on medical imaging features.


Modeling, Parallel Computing


IM/TH- Foundational Skills: Hardware acceleration and parallel processing

Contact Email