Dr. Zheng XiaHe originally aimed for a career in engineering.
His dive into cancer biology came after an experience working on a machine learning project at Houston Methodist, an academic medical center in Texas. There, he spent every day in the hallways seeing sick patients and worried families.
Although his training had nothing to do with biology or cancer, “I felt that the effort to go into this field was very worthwhile,” he says.
The year was 2007, and Shea had heard about rapid advances in gene reading technology, including microarrays that can measure the activity levels of thousands of genes at once. The torrent of data from such experiments overwhelmed traditional biological analysis methods.
Xia was confident that his expertise could help cancer researchers pinpoint the answer. He was right.
Currently, Xia is an associate professor of biomedical engineering in the OHSU School of Medicine and a member of the OHSU Knight Cancer Institute. His lab develops bioinformatics tools, including machine learning systems, that can interpret biomedical datasets beyond the scale of direct human understanding. He and his collaborators have published multiple papers in high-impact journals including: Nature, nature biotechnologyand clinical cancer research. He is the principal investigator on three of his grants and co-investigator on nine of his major R01-level grants from the National Institutes of Health and other external funding sources.
Working collaboratively with biologists and clinicians is key.
“For me, acting on important biological or clinical questions is very important, but my background is in engineering, so I don't know the clinically important questions,” Xia said. Masu. “Through collaboration, we learn how to think about problems and how biologists formulate hypotheses. And while we answer questions that our collaborators are interested in, we also learn how those questions challenge us from different angles.” It motivates us to think of new tools for analyzing data.”
work together Dr. Amy MoranShea, an associate professor of cell biology, developmental biology, and cancer biology, helped gain more insight from single-cell analyzes of RNA, the genetic messenger molecule transcribed from active genes. . The study, led by Moran, revealed how the sex hormones androgens (most commonly testosterone in men) limit the body's response to cancer immunotherapy. This discovery could help make these therapies more consistently effective.
disease behavior
The project will allow Xia and members of his lab to focus on a subpopulation of cells within tumors that cause important disease behaviors, such as the ability to resist cancer drug treatment. This provided an opportunity to develop a calculation tool. This method expands the scope of single-cell analysis that has been limited by small sample sizes, but results in insufficient statistical power to answer some important questions about tumors. It will be.
Scissor combines single-cell findings with clinical outcome data available at the whole-tumor level in public databanks such as the Cancer Genome Atlas. In demonstrating Scissor's utility, researchers identified an aggressive cancer cell subpopulation associated with worse survival outcomes in lung adenocarcinoma tumors, the most common type of lung cancer. Although he only had single-cell data from two cancer patients in The Cancer Genome Atlas, clinical outcome data came from his 471 patient samples. Identifying which cell subpopulations are involved in drug response, tumor progression, and cancer spread could help uncover mechanisms and point the way to better targeted therapies.
Recently, Xia et al developed PENCIL. This uses single-cell gene activity data to not only select important subpopulations of cancer cells, but also to reveal subpopulations of cells that are continuously transitioning between states. It is a pioneering machine learning model that can. Cells that change from normal to malignant, etc.
PENCIL employs a strategy called “learning with rejection.” In traditional supervised machine learning, a computer model must choose labels to predict new samples from the labels provided to train the model. Learning with rejection gives machine learning systems the ability and freedom to say “I don’t know” and reject unreliable decisions.
“This is a very new idea,” says Shea. “I believe we are the first to apply rejection learning to biomedical research.”
By eliminating unrelated cancer cells and focusing on relevant cancer cells, PENCIL can recognize genetic signals missed by standard models and make more accurate predictions. Xia said this strategy is “in line with what Confucius once said: true wisdom is to say you don't know the answer to a question you're not sure about, and only answer questions you're confident in.”
“Machine learning allows anyone to make predictions, but in medicine, where our predictions can influence patient treatment decisions, we need more reliable predictions. It’s important,” he continues. “If you predict that this patient will respond to a particular drug, you should have a high degree of confidence.”
OncoGPT
Currently, Xia is working on a project he nicknamed oncoGPT, after the famous text generation artificial intelligence tool ChatGPT. ChatGPT is trained on a vast amount of publicly available text to produce consistent, grammatically correct document content.
Xia's oncoGPT is trained on a publicly available database containing gene expression data from millions of cancer cells that can be related to factors such as immune response, treatment resistance, and patient survival. It works in a similar way. This project received the 2024 OHSU Faculty Excellence and Innovation Award, sponsored by the Silver Family Innovation Fund.
“This will be a foundational model to facilitate unique clinical data analysis where sample size may be limited,” he says.
For example, this model can be used in the Knight Cancer Institute's SMMART program to predict which targeted therapy drug or combination of drugs will be most effective for a patient. In the SMMART program, researchers orchestrate multiple technologies to closely analyze each person's tumor and how to track it. Cancer cells evolve over time in response to treatment.
Given the breathtaking pace of advances in artificial intelligence, Xia is both excited about the opportunity and nervous about keeping up. But he says: “At the Knight Cancer Institute, he feels there has never been a better time to do big data analytics and machine learning for translational cancer research.”
