When Dr. Xiyu Peng opens an image of a tumor tissue slide on her screen, she doesn’t see a jumble of colors. She sees stories, millions of cells working together, forming intricate patterns that may hold the key to defeating cancer.
Now, thanks to her computational tool SpatialTopic, the Texas A&M University statistician is helping scientists worldwide read those stories with clarity and speed.
Modern imaging technologies have transformed cancer research. A single tissue slide can contain millions of cells, capturing everything from protein markers to cellular structure. Yet the real challenge lies not in collecting that information but in interpreting it.
“That was the problem we faced,” Peng recalls. “We had all this incredibly rich information in tissue samples from cancer patients, but there was no good way to analyze it at scale.”
Her answer was SpatialTopic, a fast, scalable, and easy-to-use tool that decodes how cells are arranged in complex tissues. The research, published in Nature Communications, marks a significant advance in understanding how diseases such as cancer progress and respond to treatment.
With SpatialTopic, scientists can detect key spatial patterns such as clusters of immune cells that reveal how the body fights tumors without requiring manual review by a pathologist. The program can process data from about 100,000 cells in under a minute on a standard laptop, enabling analyses that were once unthinkable.
From Memorial Sloan Kettering to Aggieland
The idea for SpatialTopic began at Memorial Sloan Kettering Cancer Center (MSKCC), where Peng worked as a biostatistician before joining Texas A&M in 2024. At MSKCC, she helped a team analyzing imaging data from a new imaging platform that can examine dozens of protein markers simultaneously on a single tumor slide. This technology, spatial proteomics, is so groundbreaking that Nature Methods named it “Method of the Year 2024.”
“In 2021, there wasn’t much literature about how to analyze this type of data,” Peng says. “We had to build everything from the ground up.”
Her fascination with topic modeling, a statistical method originally used to find themes in text, sparked the idea of applying it to spatial proteomics. By treating each tissue region as a “document” and cells as “words,” she could uncover recurring spatial patterns, “topics”, or tissue motifs across samples.
Discovering Cancer’s Hidden Signatures
When Peng and her collaborators reviewed hundreds of melanoma slides, they noticed clusters of immune cells that consistently appeared in patients who responded well to immunotherapy. After consulting medical oncologists, they confirmed these clusters were tertiary lymphoid structures, or TLSs, specialized immune formations that help the body fight tumors.
Their findings, published earlier this year in Cell Reports, showed TLSs to be powerful biomarkers for melanoma immunotherapy treatment response. In their following works, it is demonstrated how SpatialTopic could automatically detect interpretable and clinically relevant spatial patterns, like TLS, from large imaging datasets.
“TLSs are one of the hottest topics in immuno-oncology,” Peng says. “Everyone in this field is excited about how computational tools like ours can reveal them objectively and consistently.”
Today, her collaborators at MSKCC use SpatialTopic to analyze datasets containing more than 40 million cells across 100 tissue slides on modest servers, thanks to its efficiency.
What sets SpatialTopic apart, Peng says, is not only its scientific precision but its accessibility. The analysis can be run with a single line of code, and the method works across both spatial proteomics and transcriptomics data when cell types are known.
“I spent a lot of time improving the scalability of our R package in both speed and memory usage,” she says. “My goal was to make this tool something any researcher can use, even without high-end computing resources.”
That philosophy of democratizing bioinformatics drives much of Peng’s work. She hopes to place advanced analytical power directly in the hands of biologists, clinicians, and data scientists.
Nurturing Bioinformatics at Texas A&M
For Peng, Texas A&M offered the ideal home to expand this vision. “The Department of Statistics provides such a supportive environment for bioinformatics research,” she says. “We have excellent computational infrastructure and seed funding that encourages collaboration across departments like biology and biomedical sciences.”
She is particularly excited about the university’s new undergraduate major in bioinformatics, which she believes will prepare the next generation of researchers at the intersection of data and biology. “I can’t wait to mentor students who are as passionate about this field as I am,” she says.
Her path to bioinformatics began with that same curiosity. As an undergraduate, Peng double-majored in biotechnology and applied mathematics, a combination that foreshadowed her career. Her Ph.D. adviser, Dr. Karin Dorman, introduced her to statistical modeling for complex biological systems, and Peng realized she could use data science not just to describe the world but to change it.
Statistics allows us to translate patterns in nature into knowledge. It’s the bridge between what we observe and what we understand.
Looking Ahead
At Texas A&M, Peng is building a research program that partners with leading medical centers including Memorial Sloan Kettering, Mayo Clinic, UConn Health, and the Providence Institute. She is expanding SpatialTopic to handle hundreds of tissue images for population-scale cancer studies and to identify new biomarkers that could guide clinical decisions.
Within her department, she collaborates with Dr. Huiyan Sang to develop advanced statistical and computational methods for large-scale spatial molecular data. Together, they aim to push the limits of what big data can reveal about human health.
“I believe I’m working in a field that truly makes a difference,” Peng says with a quiet smile. “Every dataset we analyze brings us one step closer to improving lives.”
.