Agent-based Modeling of Biological Active Matter
Restricted (Penn State Only)
- Author:
- Li, Changhao
- Graduate Program:
- Engineering Science and Mechanics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 23, 2024
- Committee Members:
- Sulin Zhang, Chair & Dissertation Advisor
Qing Wang, Outside Unit Member
Adri van Duin, Outside Field Member
Albert Segall, Program Head/Chair
Amir Sheikhi, Outside Field Member
Nanyin Zhang, Major Field Member - Keywords:
- Biophysics
Applied Mechanics
Morphogenesis
Self-patterning
Scientific Machine Learning
Agent-based Modeling - Abstract:
- A hallmark of active matter is its ability to metabolize free energy to create complex collective motions and patterns varying in space and time. Active matter systems are composed of self-driven agents capable of exchanging energy, mass, and momentum with the surrounding environments, converting internal and external free energy into directed movements. The active agents, the characteristic length scale of which ranges from subcellular to macroscale, often possess diverse inter-agent interactions and communications, giving rise to rich phenomena of out-of-equilibrium behavior, such as collective motions, phase separations, and morphogenesis. Therefore, mechanics plays a crucial role in deciphering intricate interactions at different lengths and time scales. This dissertation focuses on the mechanics of active matter, especially on self-patterning and morphogenesis of several specific systems, including mechanically confined biofilms, microgels-fibroblasts mixtures, and epithelial cell monolayers. Targeting at addressing how those active agents interact with the surrounding environment and themselves, biophysical models are developed using the methodologies of agent-based modeling, continuum theories, and machine learning. The content of this dissertation is divided into three parts: stress-mediated cell ordering and shape bifurcation in growing biofilms under confinements, aggregation dynamics of microgel-fibroblast spheroids, and data-driven modeling of cellular traction force maps. As a start, the first part strives to explore the biomechanics of how V. cholerae biofilms develop into distinct shapes and internal cell orderings inside hydrogels with different stiff- nesses. A minimal agent-based model is developed, which not only quantitatively reproduces the biofilm growing process, but also reveals the effect of mechanical stresses where rod-like cells tend to align with the local minimal compression direction. The model highlights the growth-induced stress history, which is difficult to track through experiments. Meanwhile, the simulation package for the agent-based model is a powerful and highly extensible tool for active matter research. The second part focuses on the aggregation process of biohybrid spheroids composed of fibroblast cells and adhesive microgels. The tailored agent-based model recapitulates an interesting 2D to 3D structural transition during the self-assembly process of spheroids, which is powered by the integrin-mediated cell-microgel adhesion and active self-propulsion of fibroblast cells. Cells actively crawl on the substrate and attach to microgel surfaces, then gradually self-assemble into porous tissue-like structures, where the assembling dynamics exhibits a triphasic characteristic. Our agent-based model and kinetic theory quantitatively address the relation between cellular active forces, cell-gel adhesion, and the triphasic kinetics of aggregation. In the final part, A machine-learning-based approach is established to predict traction force maps of contractile cell monolayers. Conventional traction force microscopy suffers from shortcomings such as low efficiency and high cost, requiring tedious effort from experienced technicians. Meanwhile, the biophysical origin of cellular traction forces has been well-studied. Resorting to rapidly developing machine learning algorithms, we combine experimental traction force microscopy and continuum modeling to build up a large dataset of traction force maps and train a generative adversarial network. The machine learning model exhibits high fidelity and remarkable efficiency, which only requires the deformed profiles of cell monolayers and a few other biophysical parameters. The technique serves as a strong candidate estimating cellular traction forces at a low cost, providing a data-driven framework to address other complex biophysics problems.