As an industrial data science researcher, Kaibo Liu has devoted much of his adult life to improving quality in manufacturing systems and monitoring the health of machines.
That changed in 2021, when the professor of industrial and systems engineering at the University of Wisconsin-Madison expanded his focus to human health.
Liu’s then-9-year-old son, Joseph, was diagnosed with a cancerous germ cell tumor. Inspired by watching his son’s treatment at St. Jude Children’s Research Hospital in Memphis, Tennessee, while on sabbatical, Liu resolved to give back through his work.
“When we talk about human life, nothing is more important than that,” says Liu. “If I can save people’s lives, if my research can contribute only a small percent to the better life of human beings, I think my research is worth it.”
Joseph (right) and Kaibo Liu
His subsequent efforts have started to bear fruit. Liu and collaborators at St. Jude have published a paper in the journal Nature Methods showcasing a first-of-its kind algorithm in the field of spatial transcriptomics. The algorithm uses generative artificial intelligence and allows investigators to characterize gene expression in tissues down to the single-cell level. The researchers have made their tool available for free on GitHub.
The field of spatial transcriptomics includes several methods that can reveal gene expression within a tissue sample—information that’s useful for determining which genes are at work in a disease, better understanding disease progression, and informing treatments.
However, the leading methods either can’t break down data to single-cell resolution or can’t cover the full genome. Those limitations offered an opportunity when Liu connected with St. Jude faculty member Jiyang Yu, a computational biologist.
Adapting methods they’ve used in industrial applications, Liu and former PhD student Ziqian Zheng built a type of statistical model called a generative probabilistic model to create a tool they’ve dubbed “Spotiphy.” The tool can determine the proportion of each type of cell within a given location and then break down aggregated gene expression data for the region to the single-cell level.
Ziqian Zheng
“One major advantage of our method is we go one step beyond getting to know the cell types in each tiny area,” says Zheng (PhDIE ’24), now a software engineer at the autonomous driving tech company WeRide. “Knowing which cells are there is pretty helpful, but if we can get to know the gene expression of each cell, it will be more helpful to the biologists.”
Because it’s trained on large collections of both spatial transcriptomics data and tissue images, Spotiphy can fill in gaps in the readout of gene expression.
“Imagine a picture of a hand, but the middle is missing,” says Junmin Peng, a faculty member at St. Jude and one of the paper’s senior authors. “The algorithm has acquired general rules from its training so that it can impute the missing part of the picture—like reconstructing where and what the palm should look like, or in reality, the space between imaging spots.”
The researchers validated the tool by analyzing mouse models of Alzheimer’s disease and cancerous tissues.
Three years after diagnosis, Joseph Liu’s cancer is in remission, thanks to the treatments he received at St. Jude. The hospital covers all treatment costs through donations, meaning patients don’t receive a bill.
Kaibo Liu hopes to keep paying back the larger medical community through his work. In partnership with their St. Jude collaborators, Liu and Zheng continue to refine their Spotiphy tool and hope to expand its capabilities.
“I feel so grateful,” Liu says. “That’s why I really wanted to do something, if I could, to return back to this amazing hospital and to this research work.”
Kaibo Liu is the Grainger STAR Professor of industrial and systems engineering. Jiahui Zhang, a PhD student in industrial and systems engineering, is the other UW-Madison author on the paper.
This research was funded by the National Institutes of Health (award numbers R01GM134382, U01CA264610, U01CA281868, R01CA274251 and RF1AG068581) and ALSAC, the fundraising and awareness organization of St. Jude.