Transforming complex data into clarity

Long before any equations or algorithms come into play, Jingyi (Ginny) Zheng thinks about people. Every dataset she works with — whether MRI scans, cardiac images, gene expression data or photos of household pets — represents a real question someone needs answered. For a cardiologist, it may be whether a patient needs surgery before symptoms worsen. For a neuroscientist, it may be how changes in brain connectivity relate to neurological conditions. For collaborators in the College of Veterinary Medicine, it may be which puppies show early signs of excelling in detection work.

Those human questions sit at the center of Zheng’s research and guide the statistical and machine learning methods she develops.

When Zheng joined Auburn in 2019, she arrived as a freshly minted PhD with no built-in research network. She began introducing herself to anyone who might need a data scientist. It didn’t take long for her to spot a pattern. Every collaborator had data, often more than they could ever analyze. What they needed was someone who could help turn it into something useful.

"We have so much data now, but making it useful is the important part," she said. "We need expertise to translate data into knowledge. That is the key part of our work."

Zheng, an associate professor in the Department of Mathematics and Statistics and the College of Sciences and Mathematics’ 2025 Dean’s Faculty Research Award winner, is a statistician by training. But her research goes far beyond applying existing methods. Her work centers on developing new statistical and machine learning techniques, with collaborations guiding the creation of tools that can generalize across fields.

Her collaboration with cardiologists at UAB is one of her longest running. Together they follow patients with cardiac disease every six months for two years, collecting MRI scans, echocardiograms, blood samples and other biomarker data.

"We build models to predict postsurgical performance," she said. "Some patients do not have symptoms and do not realize they have heart problems. Others have mild symptoms but not enough for surgery. We want to know who is at risk before it gets worse."

The work requires patience. Zheng had to learn cardiology vocabulary, and the physicians had to learn how prediction models work.

"Every field has its own language," she said. "I enjoy being the bridge, translating complex data into meaningful insight."

Over time, the two sides built a shared language and introduced deep learning tools that analyze MRI images far faster than manual tracing.

Zheng has carried those techniques into other partnerships. With Auburn’s MRI Research Center, she helps convert brain scans into networks that map how different regions communicate. Those patterns can differ in conditions such as autism, Alzheimer’s disease and PTSD.

"If the brain has issues, some connections break," she said.

She added that by comparing those connectivity maps, her group is developing machine learning tools that could support diagnosis.

Because her work focuses on methodology, Zheng often finds that techniques from one project open doors in others. Image-based models first developed for MRI data now support computer vision tasks. Beyond medicine, she collaborates with engineers who work with signals, CT scans and fatigue performance data, and with chemists and pharmacists whose datasets require complex analysis tools.

"The most exciting part is developing tools that work across domains," Zheng said.

One of her most unusual collaborations involves working dogs trained for detection. The project spans genetics, microbiomes, gene expression and performance evaluations from puppyhood through training.

"We analyze their data to see how performance links to final outcomes," she said.

The goal is to understand which traits predict success and how early selection could make training more efficient.

She is also working with veterinarians to develop tools for pet owners. Using images collected from veterinary medicine students and volunteers, her models estimate a cat’s body condition score. In the future, Zheng’s team expects to use the model in telemedicine portals so users can get an estimation of their cat’s condition score.

"Owners always say their cat is a healthy weight," she said. "But a model can give unbiased feedback. If the score is high, it may be time for a vet visit."

Across every project, Zheng sees the same pattern. Her collaborators are experts in their fields and understand the scientific questions, but they often cannot extract everything their data contains.

"They pick out the information they need and leave the rest behind," she said. "We try to use the full dataset and find more useful information."

For Zheng, collaboration is the engine of her research. Each partnership teaches her something new and broadens the impact of the tools she develops.

"Collaboration makes my life more colorful," she said. "There is always more we can learn from each other and more knowledge hidden in the data."