Next Scientifically Speaking session focuses on machine learning and cognitive function

Register here to attend the October 10 virtual event

The Jackson Laboratory's virtual series 'Scientifically Speaking' continues October 10 with Jaycee Choi, a predoctoral associate who is using machine learning tools to study the behavior and cognitive function of mice in their natural state. She hopes her work will improve our therapeutic approach to metabolic disease, neuropsychiatric conditions, Alzheimer's disease and more.

Choi came to JAX straight from her undergraduate studies at La Sierra University in Riverside, Calif. Though she graduated with a biology degree, she started as a pre-med English major hoping to focus on creative writing.

"Growing up, I really loved writing and reading," she said. "I thought science was super rigid, all factual, just a procedure that you learn to follow. It wasn't until halfway through college that I started getting interested in research and realized you can be creative in science as well."

Fast, vast data

In the laboratory of Vivek Kumar, Ph.D., Choi is exploring homeostatic behaviors - basic activities like eating, drinking and sleeping that keep the body regulated. She uses machine-learning-enabled video recordings to collect data on these behaviors faster than a human ever could.

"Let's say we're recording three mice at 30 frames per second, 24 hours day over the course of four days," Choi said. "On top of that, we conduct dozens of experiments with each mouse per day. The analysis wouldn't be possible for a human. With machine learning, we feed all that data into an algorithm and get results in seconds."

For example, Choi can compare different mouse strains for patterns in the amount of food they consume and how their feeding relates to their sleep cycle. These results could provide insight on metabolic diseases such as obesity and diabetes. For these and other health conditions, data helps turn encouraging research into candidates for preclinical and clinical testing[PM1].

Letting mice be mice

Choi's work advances one of the Kumar lab's primary goals: to streamline the research pipeline by limiting human interference in the testing environment, allowing mice to behave more naturally and yielding more accurate results. She hopes her work alleviates last-minute environmental factors that cause drugs to fail in human trials.

"Often you have a drug that you think works and gets all the way through the preclinical pipeline at no small cost, and then it fails in a human clinical trial," Choi says. "We want to mitigate that issue by getting results in a more naturalistic environment."

The maze test and Alzheimer's disease

Choi has applied this approach to a maze assay she has developed to test cognitive function in mice. She is particularly interested in finding better ways to detect early-onset Alzheimer's disease, and in how the disease manifests differently in males and females.

Her assay is promising because of its hands-off quality: Mice choose whether to enter it, and Choi can observe their behavior as it naturally unfolds. She hopes the elimination of stressors (such as water tests or touchscreen assays, on which mice must be trained for months) will help better interpret their behavior.

"We tend to assume that mice have some sort of internal state," she said. "For example, we use mice to study anxiety, but in the end, you can't ask a mouse if it's anxious. You have to look at their behavior and develop new ways of quantifying it and correlating it to an internal state we can never truly measure."