Velda Wang is a co-lead author on the paper in Brain Sciences
Velda Wang started looking for opportunities to dig into the human brain the summer after her sophomore year at Parkview High School. While taking an Advanced Placement biology class, she’d become interested in Alzheimer’s disease, discovered there wasn’t a cure or known cause, and decided she wanted to help change that.
Wang soon found her chance in the lab of Cassie Mitchell, assistant professor in the Wallace H. Coulter Department of Biomedical Engineering at Georgia Tech and Emory University.
“I wanted to get involved in research, delve deeper than I could in high school, and see if I could make an impact, even in a small way,” Wang said. And she did — Wang is co-lead author of a peer-reviewed academic paper published in the journal Brain Sciences, the result of her work as a high-school intern in Mitchell’s research group.
In part, the paper describes how Wang designed an algorithm that could give researchers new avenues to pursue when probing why certain drugs or supplements give a person risk or resilience to Alzheimer’s. And it all started with a cold email to Mitchell in the summer before Wang’s junior year of high school.
“I reached out to a number of professors and was so grateful when Dr. Mitchell replied and agreed to take me on as an intern,” said Wang, who now is studying neuroscience at Duke University with an eye toward medical school. “I’m also grateful for the mentorship I received in the lab, and for the opportunity to lead a project. That was a real honor, and the experience solidified my interest in science. It’s why I’m pursuing my degree in science now.”
While many of her high school classmates were making plans for spring break, Wang was putting finishing touches on her study using machine learning to examine the impact of different drugs on different sub-populations of Alzheimer’s disease patients. It was published in July.
“The draft she wrote was first-class, graduate-level writing,” Mitchell said. “We’ve had some talented high school students working in our lab. But Velda actually took the lead role and authorship in a research project, and that hasn’t happened before.”
Mitchell said her Laboratory for Pathology Dynamics has hosted about 45 high school interns over the past five years in a program she developed to serve a few purposes.
“It’s getting more competitive to get into colleges these days, even with good grades. So giving students opportunities to shine beyond their high school classrooms is important, I think, in helping to develop a new generation of confident scientists,” Mitchell said. Plus: “We need a lot of help in my lab, and these are talented students.”
Mitchell’s lab uses big data, machine learning, biostatistics, and bioinformatics to identify the causes of disease, predict new therapeutics, and optimize or repurpose drugs that already are available. The research mostly targets neurological disease but also has applications cancer, pediatrics, cardiovascular medicine, and other areas.
That means there’s plenty of work for young researchers — lots of keystrokes and reading, literature review and data curation. For the more ambitious, like Wang, those tasks can progress to writing code, making algorithms, and leading a project.
From the beginning, Wang said she wanted to dig into the research. There was just one problem: Mitchell’s research program is heavy on computer science, and Wang had no computer science experience. So, she learned.
“I had to teach myself a little bit,” Wang said. “Just enough to do the project.”
It was a bit more complicated than that, according to Mitchell. Wang taught herself Python, a common programming language used in machine learning.
She shared the lead author role for the Brain Sciences study with fellow researcher Jayant Prakash. He graduated with his master’s degree in computer science while they were working on the study, leaving Wang to guide the project to completion alongside undergraduate research assistant Robert Quinn, the paper’s other co-author (with Mitchell).
The research aimed to tackle a key issue with Alzheimer’s disease: Patients have vastly different attributes (like genetic backgrounds, lifestyles, or pre-existing conditions), onset age and symptoms, and progression patterns.
“This is problematic for clinical trials,” Mitchell said. “It means one drug might have a really good chance of helping a certain Alzheimer’s disease sub-population with similar traits, but if we measure that drug across the general disease population in a trial, it dilutes the effect such that we may discard a drug that would otherwise be really helpful for a sub-population of patients.”
The researchers used machine learning to “cluster” patient sub-populations. Prakash led that part of the project. Then they used a type of machine learning called association rule mining, or ARM, which examined the impact of drugs and supplements on different sub-populations. That was Wang’s part.
“She made the ARM algorithm for drug mining, then she had to map those drug patterns back to the sub-population clusters Jayant found with his algorithms,” Mitchell said.
It’s exciting work, she believes, because the sub-population clusters they identified will generalize to other Alzheimer’s patient populations, “and their attributes will help future trials have some guidance on how Alzheimer’s patients could be segregated to help find drugs that work for a least a portion of patients.”