Ten fruit flies buzz about a petri dish in Kristin Scaplen Ph.D.’s lab while a camera, positioned above them, records their every movement. For the next 20 minutes, Christian Khoury ’27 — who built a machine learning model to collect data on the insects — tracks how the flies interact with each other, their turning rate, and their sedation phases. The student researcher’s goal is to understand the social and locomotor changes that occur in the fruit flies when they are exposed to alcohol.
Khoury, a Finance major with a double minor in Psychology and Political Science, has spent the past two years teaching himself machine learning, including how to create vision models (AI systems that are designed to process and understand visual data) and how to use segmentation (the process of dividing data into meaningful subgroups to improve analysis, model performance, and pattern recognition).
Having published his own machine learning models online and worked with several startups, Khoury is also the founder of Deep Autonomy — a company that develops AutoML frameworks that build, train, and validate AI models with minimal human supervision. Currently developing models that segment adult glioma, a type of brain tumor, through MRI scans, Khoury’s growing love for neuroscience led him to Scaplen’s lab this past summer.
“I thought I could really improve my skills and test myself while also building something that her lab would need for the future,” Khoury says.
RELATED ARTICLE: Unlocking the mind's secrets
The idea of working in Scaplen’s lab came to fruition after Khoury took her “Introduction to Neuroscience" course and found himself intrigued by the area of study. Asking if she had any available summer positions in her lab, the two met and discussed his background as well as the research that her lab is devoted to. Khoury learned that Scaplen and student researchers use fruit flies to gain insight on the mechanisms in the brain that underlie how maladaptive memories are encoded. By studying how these memories are formed, they hope to better understand why people develop addictive or other unhealthy behaviors.
Looking to build on the lab’s existing tracking capabilities, Khoury’s new machine learning model gives current, and future, researchers an updated tool to monitor and analyze the flies' behaviors during experiments. Breaking down how his two-segment model works, Khoury explains that the first portion segments the fly while the second one segments the fly’s body into three parts: head, thorax, and abdomen.
“From that, we're able to look at how they interact with other flies in the dish,” he says.
Khoury, who used Python to code the model, leaned on You Only Look Once (YOLO), a series of real-time object detection systems, for his model’s framework.
“YOLO is easy to set up, and it works extremely well,” says Khoury, adding that it can take several days to configurate. “There's a lot of things that go into it: You have to collect data, you have to annotate it, then you have to train it, and run inference on it.”
When the experiments get rolling, the first five minutes are devoted to tracking the flies at their baseline when they haven’t been given any alcohol. From minutes five through 10, the fruit flies experience some alcohol exposure, before having more exposure during the next five minutes. From minute 15 to 20, Khoury watches them enter their recovery phase.
Having wrapped up his internship, Khoury is now writing a research paper on the project with the hopes of publication. He shares that working in Scaplen’s lab has been a rewarding and fun experience — as well as a big learning curve.
“It's cool seeing stuff that I've worked and tried at home and applying it somewhere that makes a difference,” says Khoury, who is applying to master’s programs in neuroscience and finance. “I have a lot of gratitude for Bryant for giving me the opportunities that they do.”