Related Experiment Video
Updated: May 11, 2026
10:25
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
8.4K
Surrogate modeling of Cellular-Potts Agent-Based Models as a segmentation task using the U-Net neural network
Tien Comlekoglu1,2, J Quetzalcóatl Toledo-Marín3,4, Tina Comlekoglu5
1Department of Biomedical Engineering, University of Virginia, Charlottesville, VA, USA.
Arxiv
|May 19, 2025
Summary
This study introduces a deep learning surrogate model to speed up complex Cellular-Potts Model (CPM) simulations. The U-Net convolutional neural network accelerates biological process modeling by 590x, capturing emergent behaviors like vasculogenesis.
Area of Science:
- Computational Biology
- Biophysics
- Machine Learning
Background:
- Cellular-Potts Models (CPMs) are widely used for simulating multicellular biological systems.
- CPMs are computationally intensive due to agent interactions and partial differential equations (PDEs).
Purpose of the Study:
- To develop a deep learning surrogate model to accelerate CPM simulations.
- To investigate the potential of U-Net convolutional neural networks (CNNs) for this purpose.
Main Methods:
- Developed a CNN surrogate model with a U-Net architecture.
- Trained the model to predict 100 Monte-Carlo steps (MCS) ahead.
- Applied the model to accelerate simulations of *in vitro* vasculogenesis.
Main Results:
- Achieved a 590x acceleration in simulation evaluation compared to standard CPM execution.
- The surrogate model accurately captured emergent behaviors, including vessel sprouting, extension, anastomosis, and lacunae contraction.
- Demonstrated effective recursive evaluation of CPM simulations.
Conclusions:
- Deep learning, specifically U-Net CNNs, can serve as efficient surrogate models for computationally expensive CPM simulations.
- This approach enables faster evaluation of biological processes at larger spatial and temporal scales.
- Facilitates advanced research in areas like developmental biology and tissue engineering.
