Shayn Peirce-Cottler is a Pioneer of Agent-based Modeling

Professor Shayn Peirce-Cottler fashions virtual worlds that grow and develop over time. Like a video game designer, she uses preset rules that govern the interaction of actors and their environment. Alter the environment or modify the character of an individual actor, and the outcomes change.

Of course, game designers are free to create their own worlds. Peirce-Cottler sets herself a more exacting task — replicating as much as possible the world within the body. Through the use of agent-based modeling, Peirce-Cottler hopes to shed light on how complex webs of gene expression, chemical signaling, and cell behavior contribute to human development as well as disease.

Peirce-Cottler — a professor in the UVA Biomedical Engineering Department who was recently named president-elect of the Microcirculatory Society, an organization that promotes innovative research and teaching to increase the understanding of microcirculatory function in health and disease — is currently using agent-based modeling to study changes in blood vessels during diabetes and adaptations in muscle due to injury, but its potential applications in biology are limitless.

As one of the technique’s pioneers, this is a point she often makes during the many invited presentations she gives on its advantages. In the last 18 months alone, she has spoken to audiences at MIT, Yale, UT Austin, and the NIH among other venues. Peirce-Cottler was also invited to serve on the NIH’s Modeling and Analysis of Biological Systems study section.

“Even five years ago, the section would see only two or three proposals a year involving agent-based modeling,” she says. “Now we regularly see 10 or so proposals that incorporate some aspect of it. It’s exciting to see more and more researchers using this modeling approach.”

Virtual Experimentation

Although agent-based modeling ultimately requires knowledge of complex mathematics and advanced computation, its successful deployment rests on a fundamental skill — careful reading. Defining rules for agents requires an exhaustive review of the scientific literature. “We pour over hundreds of papers that document what I call reductionist experiments,” she says. “That is where researchers have measured a particular behavior of a cell in response to a well-controlled environment.”

Once Peirce-Cottler and her team formulate a series of rules and combine them into a model, they subject them to a series of tests. Oftentimes, their reading uncovers contradictory findings. They can translate each of the findings into an algorithm and insert them, one at a time, into their model to determine which one best matches the biology.

They can also conduct sensitivity analyses, systematically modifying or eliminating a single rule to determine how it affects their virtual world. “One of the benefits of agent-based modeling is that you can determine in a high throughput, rigorous, and systematic manner which element in a complex web of signals is driving system behavior,” she says.

Over time, Peirce-Cottler’s team refines and enlarges their models strategically, for instance adding rules for a signaling pathway that might be altered by a specific drug or class of drugs. Their goal is to create a better representation of the biology or to ask a new question. At the same time, they can add algorithms that add more nuance to their depiction of the cell’s environment. They can take a number of approaches, replicating the mechanical environment, for instance, or the chemical environment.

Although the primary use for the models is to guide their experimental research, the models are valuable in themselves. Peirce-Cottler regularly makes them available online. “These models can save researchers an enormous amount of time and money,” she says. Her group conducted an analysis of a model that required two nights of computer time to run. Without the model, they estimate they would have had to conduct approximately 600 experiments each taking 21 days.

The efficiency and power of agent-based modeling are the primary reasons it has found its place among the tools that department researchers use. To cite two examples, Associate Professor Kevin Janes’s lab recently published an agent-based model of tumor growth and Professor Jeffrey Holmes published an agent-based model of scar tissue formation and cardiac disease.

Taking the Next Step

Peirce-Cottler has already taken on one of next big challenges in agent-based modeling — the transition to multiscale models. She is collaborating with Associate Professor Silvia Blemker to link agent-based models of muscle cells with Blemker’s finite element models of entire muscles. “By getting our models to talk to each other, we hope to gain a more comprehensive understanding of Duchenne muscular dystrophy,” she says. “Biology operates at different scales. We’re trying to replicate that.”